Simplifying a prognostic model: a simulation study based on clinical data

Feature Identification Using Interpretability Machine Learning Predicting Risk Factors for Disease Severity of In-Patients with COVID-19 in South Florida

Objective: The objective of the study was to establish an AI-driven decision support system by identifying the most important features in the severity of disease for Intensive Care Unit (ICU) with Mechanical Ventilation (MV) requirement, ICU, and InterMediate Care Unit (IMCU) admission for hospitalized patients with COVID-19 in South Florida. The features implicated in the risk factors identified by the model interpretability can be used to forecast treatment plans faster before critical conditions exacerbate. Methods: We analyzed eHR data from 5371 patients diagnosed with COVID-19 from South Florida Memorial Healthcare Systems admitted between March 2020 and January 2021 to predict the need for ICU with MV, ICU, and IMCU admission. A Random Forest classifier was trained on patients' data augmented by SMOTE, collected at hospital admission. We then compared the importance of features utilizing different model interpretability analyses, such as SHAP, MDI, and Permutation Importance. Results: The models for ICU with MV, ICU, and IMCU admission identified the following factors overlapping as the most important predictors among the three outcomes: age, race, sex, BMI, diarrhea, diabetes, hypertension, early stages of kidney disease, and pneumonia. It was observed that individuals over 65 years ('older adults'), males, current smokers, and BMI classified as 'overweight' and 'obese' were at greater risk of severity of illness. The severity was intensified by the co-occurrence of two interacting features (e.g., diarrhea and diabetes). Conclusions: The top features identified by the models' interpretability were from the 'sociodemographic characteristics', 'pre-hospital comorbidities', and 'medications' categories. However, 'pre-hospital comorbidities' played a vital role in different critical conditions. In addition to individual feature importance, the feature interactions also provide crucial information for predicting the most likely outcome of patients' conditions when urgent treatment plans are needed during the surge of patients during the pandemic.

The HEARTRISK6 Scale: Predicting Short-Term Serious Outcomes in Emergency Department Acute Heart Failure Patients

Background

Acute heart failure (AHF) is a common emergency department (ED) presentation that may have poor outcomes but often does not require hospital admission. There is little evidence to guide dispositional decisions.

Objectives

The authors sought to create a risk score for predicting short-term serious outcomes (SSO) in patients with AHF.

Methods

We pooled data from 3 prospective cohorts: 2 published studies and 1 new cohort. The 3 cohorts prospectively enrolled patients who required treatment for AHF at 10 tertiary care hospital EDs. The primary outcome was SSO, defined as death <30 days, intubation or noninvasive ventilation (NIV), myocardial infarction, or relapse to ED <14 days. The logistic regression model evaluated 13 predictors, used an AIC-based step-down procedure, and bootstrapped internal validation.

Results

Of the 2,246 patients in the 3 cohorts (N = 559; 1,100; 587), the mean age was 77.4 years, 54.5% were male, 3.1% received intravenous nitroglycerin, 5.2% received ED NIV, and 48.6% were admitted to the hospital. There were 281 (12.5%) SSOs including 70 deaths (3.1%) with many in discharged patients. The final HEARTRISK6 Scale included 6 variables: valvular heart disease, tachycardia, need for NIV, creatinine, troponin, and failed reassessment (walk test). Choosing HEARTRISK6 total-point admission thresholds of ≥1 or ≥2 would yield, respectively, sensitivities of 88.3% (95% CI: 83.9%-91.8%) and 71.5% (95% CI: 65.9%-76.7%) and specificities of 24.7% (95% CI: 22.8%-26.7%) and 50.1% (95% CI: 47.9%-52.4%) for SSO.

Conclusions

Using 3 large prospectively collected datasets, we created a concise and sensitive risk scale for patients with AHF in the ED. Implementation of the HEARTRISK6 scale could lead to safer and more efficient disposition decisions.

Prospective Evaluation and Validation of the Laryngoscore and the mini-Laryngoscore

OBJECTIVE

The Laryngoscore was described in 2014 as a practical preoperative assessment tool to predict difficult laryngeal exposure (DLE) during transoral approaches to the larynx. In 2019 the authors proposed a version with a reduced number of variables, called the mini-Laryngoscore. We aim to critically appraise and externally validate these two tools and if needed and possible, to optimize these tools.

METHODS

103 consecutive patients who underwent a microlaryngoscopy between November 2017 and June 2020 at the Leuven University Hospitals were prospectively included and subjected to a presurgical evaluation of 15 parameters and a peroperative scoring of the anterior commissure visualization. Subsequent analysis focused on the concordance of our findings with those of Piazza et al., the discriminatory ability of the test, and the validity of the included items. We then evaluated a modified prediction tool.

RESULTS

Of 103 patients, 18 (17.5%) had DLE. The Laryngoscore and mini-Laryngoscore predicted this with a "good" C-index of respectively 0.727 (95%CI: 0.608-0.846) and 0.714 (95%CI: 0.605-0.823). A newly created prediction tool including only three parameters (Interincisors gap, upper jaw dental status and previous treatments) showed a better discriminatory ability (C-index = 0.835, 95%CI: 0.726-0.944) than the original Laryngoscore, a finding that needs further external validation.

CONCLUSION

The original Laryngoscore and the mini-Laryngoscore displayed a good discriminative ability. Some parameters can be left out of the Laryngoscore without losing discrimination. An even better prediction model seems possible, using a weighted sum of selected predictor variables and by using the parameters in their continuous form.

LEVEL OF EVIDENCE

2 Laryngoscope, 134:1807-1812, 2024.

Modifiable prognostic factors of high societal costs among people on sick leave due to musculoskeletal disorders: findings from an occupational cohort study

OBJECTIVES

The objective was to identify modifiable prognostic factors of high societal costs among people on sick leave due to musculoskeletal disorders, and to identify modifiable prognostic factors of high costs related to separately healthcare utilisation and productivity loss.

DESIGN

A prospective cohort study with a 1-year follow-up.

PARTICIPANTS AND SETTING

A total of 549 participants (aged 18-67 years) on sick leave (≥ 4 weeks) due to musculoskeletal disorders in Norway were included.

OUTCOME MEASURES AND METHOD

The primary outcome was societal costs aggregated for 1 year of follow-up and dichotomised as high or low, defined by the top 25th percentile. Secondary outcomes were high costs related to separately healthcare utilisation and productivity loss aggregated for 1 year of follow-up. Healthcare utilisation was collected from public records and included primary, secondary and tertiary healthcare use. Productivity loss was collected from public records and included absenteeism, work assessment allowance and disability pension. Nine modifiable prognostic factors were selected based on previous literature. Univariable and multivariable binary logistic regression analyses were performed to identify associations (crude and adjusted for selected covariates) between each modifiable prognostic factor and having high costs.

RESULTS

Adjusted for selected covariates, six modifiable prognostic factors associated with high societal costs were identified: pain severity, disability, self-perceived health, sleep quality, return to work expectation and long-lasting disorder expectation. Depressive symptoms, work satisfaction and health literacy showed no prognostic value. More or less similar results were observed when high costs were related to separately healthcare utilisation and productivity loss.

CONCLUSION

Factors identified in this study are potential target areas for interventions which could reduce high societal costs among people on sick leave due to musculoskeletal disorders. However, future research aimed at replicating these findings is warranted.

TRIAL REGISTRATION NUMBER

NCT04196634, 12 December 2019.

Development and validation of a prognostic model for death 30 days after adult emergency laparotomy

The probability of death after emergency laparotomy varies greatly between patients. Accurate pre-operative risk prediction is fundamental to planning care and improving outcomes. We aimed to develop a model limited to a few pre-operative factors that performed well irrespective of surgical indication: obstruction; sepsis; ischaemia; bleeding; and other. We derived a model with data from the National Emergency Laparotomy Audit for patients who had emergency laparotomy between December 2016 and November 2018. We tested the model on patients who underwent emergency laparotomy between December 2018 and November 2019. There were 4077/40,816 (10%) deaths 30 days after surgery in the derivation cohort. The final model had 13 pre-operative variables: surgical indication; age; blood pressure; heart rate; respiratory history; urgency; biochemical markers; anticipated malignancy; anticipated peritoneal soiling; and ASA physical status. The predicted mortality probability deciles ranged from 0.1% to 47%. There were 1888/11,187 deaths in the test cohort. The scaled Brier score, integrated calibration index and concordance for the model were 20%, 0.006 and 0.86, respectively. Model metrics were similar for the five surgical indications. In conclusion, we think that this prognostic model is suitable to support decision-making before emergency laparotomy as well as for risk adjustment for comparing organisations.

Development and Validation of a Predictive Risk Algorithm for Bleeding in Individuals on Long-term Hemodialysis: An International Prospective Cohort Study (BLEED-HD)

Background

Individuals with kidney disease are at a high risk of bleeding and as such tools that identify those at highest risk may aid mitigation strategies.

Objective

We set out to develop and validate a prediction equation (BLEED-HD) to identify patients on maintenance hemodialysis at high risk of bleeding.

Design

International prospective cohort study (development); retrospective cohort study (validation).

Settings

Development: 15 countries (Dialysis Outcomes and Practice Patterns Study [DOPPS] phase 2-6 from 2002 to 2018); Validation: Ontario, Canada.

Patients

Development: 53 147 patients; Validation: 19 318 patients.

Measurements

Hospitalization for a bleeding event.

Methods

Cox proportional hazards models.

Results

Among the DOPPS cohort (mean age, 63.7 years; female, 39.7%), a bleeding event occurred in 2773 patients (5.2%, event rate 32 per 1000 person-years), with a median follow-up of 1.6 (interquartile range [IQR], 0.9-2.1) years. BLEED-HD included 6 variables: age, sex, country, previous gastrointestinal bleeding, prosthetic heart valve, and vitamin K antagonist use. The observed 3-year probability of bleeding by deciles of risk ranged from 2.2% to 10.8%. Model discrimination was low to moderate (c-statistic = 0.65) with excellent calibration (Brier score range = 0.036-0.095). Discrimination and calibration of BLEED-HD were similar in an external validation of 19 318 patients from Ontario, Canada. Compared to existing bleeding scores, BLEED-HD demonstrated better discrimination and calibration (c-statistic: HEMORRHAGE = 0.59, HAS-BLED = 0.59, and ATRIA = 0.57, c-stat difference, net reclassification index [NRI], and integrated discrimination index [IDI] all P value <.0001).

Limitations

Dialysis procedure anticoagulation was not available; validation cohort was considerably older than the development cohort.

Conclusion

In patients on maintenance hemodialysis, BLEED-HD is a simple risk equation that may be more applicable than existing risk tools in predicting the risk of bleeding in this high-risk population.

Rethinking Donor and Recipient Risk Matching in Europe and North America: Using Heart Transplant Predictors of Donor and Recipient Risk

BACKGROUND

In Europe, there is greater acceptance of hearts from higher-risk donors for transplantation, whereas in North America, the donor heart discard rate is significantly higher. A Donor Utilization Score (DUS) was used to compare European and North American donor characteristics for recipients included in the International Society for Heart and Lung Transplantation registry from 2000 to 2018. DUS was further evaluated as an independent predictor for 1-year freedom from graft failure, after adjusting for recipient risk. Lastly, we assessed donor-recipient risk matching with the outcome of 1-year graft failure.

METHODS

DUS was applied to the International Society for Heart and Lung Transplantation cohort using meta-modeling. Posttransplant freedom from graft failure was summarized by Kaplan-Meier survival. Multivariable Cox proportional hazard regression was applied to quantify the effects of DUS and Index for Mortality Prediction After Cardiac Transplantation score on the 1-year risk of graft failure. We present 4 donor/recipient risk groups using the Kaplan-Meier method.

RESULTS

European centers accept significantly higher-risk donor hearts compared to North America. DUS 0.45 versus 0.54, P<0.005). DUS was an independent predictor for graft failure with an inverse linear relationship when adjusted for covariates (P<0.001). The Index for Mortality Prediction After Cardiac Transplantation score, a validated tool to assess recipient risk, was also independently associated with 1-year graft failure (P<0.001). In North America, 1-year graft failure was significantly associated with donor-recipient risk matching (log-rank P<0.001). One-year graft failure was highest with pairing of high-risk recipients and donors (13.1% [95% CI, 10.7%-13.9%]) and lowest among low-risk recipients and donors (7.4% [95% CI, 6.8%-8.0%]). Matching of low-risk recipients with high-risk donors was associated with significantly less graft failure (9.0% [95% CI, 8.3%-9.7%]) than high-risk recipients with low-risk donors (11.4% [95% CI, 10.7%-12.2%]) Conclusions: European heart transplantation centers are more likely to accept higher-risk donor hearts than North American centers. Acceptance of borderline-quality donor hearts for lower-risk recipients could improve donor heart utilization without compromising recipient survival.

Whole-brain diffusion tensor imaging predicts 6-month functional outcome in acute intracerebral haemorrhage

INTRODUCTION

Small vessel disease (SVD) causes most spontaneous intracerebral haemorrhage (ICH) and is associated with widespread microstructural brain tissue disruption, which can be quantified via diffusion tensor imaging (DTI) metrics: mean diffusivity (MD) and fractional anisotropy (FA). Little is known about the impact of whole-brain microstructural alterations after SVD-related ICH. We aimed to investigate: (1) association between whole-brain DTI metrics and functional outcome after ICH; and (2) predictive ability of these metrics compared to the pre-existing ICH score.

METHODS

Sixty-eight patients (38.2% lobar) were retrospectively included. We assessed whole-brain DTI metrics (obtained within 5 days after ICH) in cortical and deep grey matter and white matter. We used univariable logistic regression to assess the associations between DTI and clinical-radiological variables and poor outcome (modified Rankin Scale > 2). We determined the optimal predictive variables (via LASSO estimation) in: model 1 (DTI variables only), model 2 (DTI plus non-DTI variables), model 3 (DTI plus ICH score). Optimism-adjusted C-statistics were calculated for each model and compared (likelihood ratio test) against the ICH score.

RESULTS

Deep grey matter MD (OR 1.04 [95% CI 1.01-1.07], p = 0.010) and white matter MD (OR 1.11 [95% CI 1.01-1.23], p = 0.044) were associated (univariate analysis) with poor outcome. Discrimination values for model 1 (0.67 [95% CI 0.52-0.83]), model 2 (0.71 [95% CI 0.57-0.85) and model 3 (0.66 [95% CI 0.52-0.82]) were all significantly higher than the ICH score (0.62 [95% CI 0.49-0.75]).

CONCLUSION

Our exploratory study suggests that whole-brain microstructural disruption measured by DTI is associated with poor 6-month functional outcome after SVD-related ICH. Whole-brain DTI metrics performed better at predicting recovery than the existing ICH score.

Predictive models for the incidence of Parkinson's disease: systematic review and critical appraisal

Numerous predictive models for Parkinson's disease (PD) incidence have been published recently. However, the model performance and methodological quality of those available models are yet needed to be summarized and assessed systematically. In this systematic review, we systematically reviewed the published predictive models for PD incidence and assessed their risk of bias and applicability. Three international databases were searched. Cohort or nested case-control studies that aimed to develop or validate a predictive model for PD incidence were considered eligible. The Prediction model Risk Of Bias ASsessment Tool (PROBAST) was used for risk of bias and applicability assessment. Ten studies covering 10 predictive models were included. Among them, four studies focused on model development, covering eight models, while the remaining six studies focused on model external validation, covering two models. The discrimination of the eight new development models was generally poor, with only one model reported C index > 0.70. Four out of the six external validation studies showed excellent or outstanding discrimination. All included studies had high risk of bias. Three predictive models (the International Parkinson and Movement Disorder Society [MDS] prodromal PD criteria, the model developed by Karabayir et al. and models validated by Faust et al.) are recommended for clinical application by considering model performance and resource-demanding. In conclusion, the performance and methodological quality of most of the identified predictive models for PD incidence were unsatisfactory. The MDS prodromal PD criteria, model developed by Karabayir et al. and model validated by Faust et al. may be considered for clinical use.

A Health Survey-Based Prediction Equation for Incident CKD

BACKGROUND

Prediction tools that incorporate self-reported health information could increase CKD awareness, identify modifiable lifestyle risk factors, and prevent disease. We developed and validated a survey-based prediction equation to identify individuals at risk for incident CKD (eGFR <60 ml/min per 1.73 m2), with and without a baseline eGFR.

METHODS

A cohort of adults with an eGFR ≥70 ml/min per 1.73 m2 from Ontario, Canada, who completed a comprehensive general population health survey between 2000 and 2015 were included (n=22,200). Prediction equations included demographics (age, sex), comorbidities, lifestyle factors, diet, and mood. Models with and without baseline eGFR were derived and externally validated in the UK Biobank (n=15,522). New-onset CKD (eGFR <60 ml/min per 1.73 m2) with ≤8 years of follow-up was the primary outcome.

RESULTS

Among Ontario individuals (mean age, 55 years; 58% women; baseline eGFR, 95 (SD 15) ml/min per 1.73 m2), new-onset CKD occurred in 1981 (9%) during a median follow-up time of 4.2 years. The final models included lifestyle factors (smoking, alcohol, physical activity) and comorbid illnesses (diabetes, hypertension, cancer). The model was discriminating in individuals with and without a baseline eGFR measure (5-year c-statistic with baseline eGFR: 83.5, 95% confidence interval [CI], 82.2 to 84.9; without: 81.0, 95% CI, 79.8 to 82.4) and well calibrated. In external validation, the 5-year c-statistic was 78.1 (95% CI, 74.2 to 82.0) and 66.0 (95% CI, 61.6 to 70.4), with and without baseline eGFR, respectively, and maintained calibration.

CONCLUSIONS

Self-reported lifestyle and health behavior information from health surveys may aid in predicting incident CKD.

PODCAST

This article contains a podcast at https://dts.podtrac.com/redirect.mp3/www.asn-online.org/media/podcast.aspx?p=CJASN&e=2023_01_10_CJN05650522.mp3.

Acute corticospinal tract diffusion tensor imaging predicts 6-month functional outcome after intracerebral haemorrhage

INTRODUCTION

Diffusion tensor imaging (DTI) can assess the structural integrity of the corticospinal tract (CST) in vivo. We aimed to investigate whether CST DTI metrics after intracerebral haemorrhage (ICH) are associated with 6-month functional outcome and can improve the predictive performance of the existing ICH score.

METHODS

We retrospectively included 42 patients with DTI performed within 5 days after deep supratentorial spontaneous ICH. Ipsilesional-to-contralesional ratios were calculated for fractional anisotropy (rFA) and mean diffusivity (rMD) in the pontine segment (PS) of the CST. We determined the most predictive variables for poor 6-month functional outcome [modified Rankin Scale (mRS) > 2] using the least absolute shrinkage and selection operator (LASSO) method. We calculated discrimination using optimism-adjusted estimation of the area under the curve (AUC).

RESULTS

Patients with 6-month mRS > 2 had lower rFA (0.945 [± 0.139] vs 1.045 [± 0.130]; OR 0.004 [95% CI 0.00-0.77]; p =  0.04) and higher rMD (1.233 [± 0.418] vs 0.963 [± 0.211]; OR 22.5 [95% CI 1.46-519.68]; p = 0.02). Discrimination (AUC) values were: 0.76 (95% CI 0.61-0.91) for the ICH score, 0.71 (95% CI 0.54-0.89) for rFA, and 0.72 (95% CI 0.61-0.91) for rMD. Combined models with DTI and non-DTI variables offer an improvement in discrimination: for the best model, the AUC was 0.82 ([95% CI 0.68-0.95]; p = 0.15).

CONCLUSION

In our exploratory study, PS-CST rFA and rMD had comparable predictive ability to the ICH score for 6-month functional outcome. Adding DTI metrics to clinical-radiological scores might improve discrimination, but this needs to be investigated in larger studies.

Modifiable prognostic factors of high costs related to healthcare utilization among older people seeking primary care due to back pain: an identification and replication study

BACKGROUND

Back pain is an extensive burden to our healthcare system, yet few studies have explored modifiable prognostic factors associated with high costs related to healthcare utilization, especially among older back pain patients. The aims of this study were to identify modifiable prognostic factors for high costs related to healthcare utilization among older people seeking primary care with a new episode of back pain; and to replicate the identified associations in a similar cohort, in a different country.

METHODS

Data from two cohort studies within the BACE consortium were used, including 452 and 675 people aged ≥55 years seeking primary care with a new episode of back pain. High costs were defined as costs in the top 25th percentile. Healthcare utilization was self-reported, aggregated for one-year of follow-up and included: primary care consultations, medications, examinations, hospitalization, rehabilitation stay and operations. Costs were estimated based on unit costs collected from national pricelists. Nine potential modifiable prognostic factors were selected based on previous literature. Univariable and multivariable binary logistic regression models were used to identify and replicate associations (crude and adjusted for selected covariates) between each modifiable prognostic factor and high costs related to healthcare utilization.

RESULTS

Four modifiable prognostic factors associated with high costs related to healthcare utilization were identified and replicated: a higher degree of pain severity, disability, depression, and a lower degree of physical health-related quality of life. Kinesiophobia and recovery expectations showed no prognostic value. There were inconsistent results across the two cohorts with regards to comorbidity, radiating pain below the knee and mental health-related quality of life.

CONCLUSION

The factors identified in this study may be future targets for intervention with the potential to reduce high costs related to healthcare utilization among older back pain patients.

TRIAL REGISTRATION

ClinicalTrials.gov NCT04261309, 07 February 2020. Retrospectively registered.

Method for Data Quality Assessment of Synthetic Industrial Data

Sometimes it is difficult, or even impossible, to acquire real data from sensors and machines that must be used in research. Such examples are the modern industrial platforms that frequently are reticent to share data. In such situations, the only option is to work with synthetic data obtained by simulation. Regarding simulated data, a limitation could consist in the fact that the data are not appropriate for research, based on poor quality or limited quantity. In such cases, the design of algorithms that are tested on that data does not give credible results. For avoiding such situations, we consider that mathematically grounded data-quality assessments should be designed according to the specific type of problem that must be solved. In this paper, we approach a multivariate type of prediction whose results finally can be used for binary classification. We propose the use of a mathematically grounded data-quality assessment, which includes, among other things, the analysis of predictive power of independent variables used for prediction. We present the assumptions that should be passed by the synthetic data. Different threshold values are established by a human assessor. In the case of research data, if all the assumptions pass, then we can consider that the data are appropriate for research and can be applied by even using other methods for solving the same type of problem. The applied method finally delivers a classification table on which can be applied any indicators of performed classification quality, such as sensitivity, specificity, accuracy, F1 score, area under curve (AUC), receiver operating characteristics (ROC), true skill statistics (TSS) and Kappa coefficient. These indicators’ values offer the possibility of comparison of the results obtained by applying the considered method with results of any other method applied for solving the same type of problem. For evaluation and validation purposes, we performed an experimental case study on a novel synthetic dataset provided by the well-known UCI data repository.

Factors influencing participants' engagement with an interactive text-message intervention to improve sun protection behaviors: "SunText" randomized controlled trial

There is growing evidence suggesting that text-message-based interventions are effective to promote sun protection behaviors. However, it is still unclear how engagement and adherence with the intervention messages can be optimized through intervention design. This study evaluated the effect of different combinations of personalized and two-way interactive messages on participant engagement with a theory-based skin cancer prevention intervention. In the SunText study conducted in February-July 2019 in Queensland, Australia participants 18-40 years were randomized to four different text message schedules using a Latin square design. This study analyzed if the order and intensity in which the schedules were received were associated with participants' level of engagement, and if this differed by demographic factors. Out of the 389 participants enrolled in the study, 375 completed the intervention period and remained for analysis. The overall intervention engagement rate was 71% and decreased from the beginning to the end of the study (82.2%-61.4%). The group starting with personalized, but not interactive messaging showed the lowest engagement rate. The intervention involving interactive messages three times a week for 4 weeks achieved the highest engagement rate. The intervention with increasing frequency (personalized and interactive three times a week for 2 weeks; then daily for 2 weeks) had lower engagement than intervention with constant or decreasing frequency. Engagement with two-way interactive messages was high across all intervention groups. Results suggest enhanced engagement with constant or decreasing message frequency compared to increasing frequency.

Predicting death in home care users: derivation and validation of the Risk Evaluation for Support: Predictions for Elder-Life in the Community Tool (RESPECT)

BACKGROUND:

Prognostication tools that report personalized mortality risk and survival could improve discussions about end-of-life and advance care planning. We sought to develop and validate a mortality risk model for older adults with diverse care needs in home care using self-reportable information — the Risk Evaluation for Support: Predictions for Elder-Life in the Community Tool (RESPECT).

METHODS:

Using a derivation cohort that comprised adults living in Ontario, Canada, aged 50 years and older with at least 1 Resident Assessment Instrument for Home Care (RAI-HC) record between Jan. 1, 2007, and Dec. 31, 2012, we developed a mortality risk model. The primary outcome was mortality 6 months after a RAI-HC assessment. We used proportional hazards regression with robust standard errors to account for clustering by the individual. We validated this algorithm for a second cohort of users of home care who were assessed between Jan. 1 and Dec. 31, 2013. We used Kaplan–Meier survival curves to estimate the observed risk of death at 6 months for assessment of calibration and median survival. We constructed 61 risk groups based on incremental increases in the estimated median survival of about 3 weeks among adults at high risk and 3 months among adults at lower risk.

RESULTS:

The derivation and validation cohorts included 435 009 and 139 388 adults, respectively. We identified a total of 122 823 deaths within 6 months of a RAI-HC assessment in the derivation cohort. The mean predicted 6-month mortality risk was 10.8% (95% confidence interval [CI] 10.7%–10.8%) and ranged from 1.54% (95% CI 1.53%–1.54%) in the lowest to 98.1% (95% CI 98.1%–98.2%) in the highest risk group. Estimated median survival spanned from 28 days (11 to 84 d at the 25th and 75th percentiles) in the highest risk group to over 8 years (1925 to 3420 d) in the lowest risk group. The algorithm had a c-statistic of 0.753 (95% CI 0.750–0.756) in our validation cohort.

INTERPRETATION:

The RESPECT mortality risk prediction tool that makes use of readily available information can improve the identification of palliative and end-of-life care needs in a diverse older adult population receiving home care.

Development and validation of a predictive algorithm for risk of dementia in the community setting

Background

Most dementia algorithms are unsuitable for population-level assessment and planning as they are designed for use in the clinical setting. A predictive risk algorithm to estimate 5-year dementia risk in the community setting was developed.

Methods

The Dementia Population Risk Tool (DemPoRT) was derived using Ontario respondents to the Canadian Community Health Survey (survey years 2001 to 2012). Five-year incidence of physician-diagnosed dementia was ascertained by individual linkage to administrative healthcare databases and using a validated case ascertainment definition with follow-up to March 2017. Sex-specific proportional hazards regression models considering competing risk of death were developed using self-reported risk factors including information on socio-demographic characteristics, general and chronic health conditions, health behaviours and physical function.

Results

Among 75 460 respondents included in the combined derivation and validation cohorts, there were 8448 cases of incident dementia in 348 677 person-years of follow-up (5-year cumulative incidence, men: 0.044, 95% CI: 0.042 to 0.047; women: 0.057, 95% CI: 0.055 to 0.060). The final full models each include 90 df (65 main effects and 25 interactions) and 28 predictors (8 continuous). The DemPoRT algorithm is discriminating (C-statistic in validation data: men 0.83 (95% CI: 0.81 to 0.85); women 0.83 (95% CI: 0.81 to 0.85)) and well-calibrated in a wide range of subgroups including behavioural risk exposure categories, socio-demographic groups and by diabetes and hypertension status.

Conclusions

This algorithm will support the development and evaluation of population-level dementia prevention strategies, support decision-making for population health and can be used by individuals or their clinicians for individual risk assessment.

Predictive scoring model for inferior alveolar nerve injury after lower third molar removal based on features of cone-beam computed tomography image

INTRODUCTION

This study aimed to construct a predictive scoring system for inferior alveolar nerve injury (IANI) following lower third molar (LM3) surgery based on cone-beam computed tomography (CBCT) images.

MATERIAL AND METHODS

Of the 1573 patients who underwent LM3 removal following the CBCT, 39 with IANI and 457 randomly selected patients without IANI were enrolled. We collected information regarding the demographic characteristics of the patients, surgical situations, and inferior alveolar canal (IAC)-related CBCT factors. The association with IANI-risk was evaluated with a backward stepwise logistic regression model as per the Akaike information criterion. Scoring models' abilities of discrimination (area under the curve) and calibration (Hosmer-Lemeshow test and calibration plots) were assessed, followed by evaluation of the clinical usefulness using decision curve analysis.

RESULTS

As per the multivariate analysis, the coronal positioned IAC on the enlarged root (odds ratio [OR], 3.78; P = 0.001), the length of perforated IAC (>3.4 mm) (OR, 3.05; P = 0.012), lingual/inter-radicular position of the IAC (OR, 3.96; P = 0.001), multiple roots closed to the perforated IAC (OR, 2.78; P = 0.025), and age >30 y (OR, 2.31; P = 0.076) were identified in the extended scoring model ranging from 0 to 12. This model was compared with our previously constructed baseline model that involved the latter three variables mentioned above, resulting in superior performance than that of the baseline model.

CONCLUSION

The extended model would be a useful tool for reliable determination of the preoperative probability of IANI.

Duration of breastmilk feeding of NICU graduates who live with individuals who smoke

BACKGROUND

Breast milk has many benefits for infants, but initiating breastfeeding/pumping can be difficult for mothers of preterm infants, especially those who smoke (or live with individuals who smoke). The primary aim of this study was to identify risks for breastfeeding/pumping cessation with neonatal intensive care unit (NICU) infants' mothers who smoke or live with individuals who smoke, using a novel survival-analytic approach.

METHODS/DESIGN

Mothers (N = 360) were recruited for a secondhand smoke prevention intervention during infants' NICU hospitalizations and followed for ~6 months after infant discharge. Data were obtained from medical records and participant self-report/interviews.

RESULTS

The sample was predominantly ethnic/racial minorities; mean age was 26.8 (SD = 5.9) years. One-fifth never initiated breastfeeding/pumping (n = 67; 18.9%) and mean time-to-breastfeeding cessation was 48.1 days (SD = 57.2; median = 30.4 [interquartile range: 6.0-60.9]). Education, length of stay, employment, race/ethnicity, number of household members who smoke, and readiness-to-protect infants from tobacco smoke were significantly associated with breastfeeding cessation. Further, infants fed breast milk for ≥4 months had 42.7% more well-child visits (p < 0.001) and 50.0% fewer respiratory-related clinic visits (p < 0.05).

CONCLUSIONS

One-quarter of infants admitted to NICUs will be discharged to households where individuals who smoke live; we demonstrated that smoking-related factors were associated with mothers' breastfeeding practices. Infants who received breast milk longer had fewer respiratory-related visits.

IMPACT

One-quarter of NICU infants will be discharged to households where smokers live. Initiating/sustaining breastfeeding can be difficult for mothers of preterm NICU infants, especially mothers who smoke or live with others who smoke. Education, employment, race/ethnicity, length of stay, household member smoking, and readiness-to-protect infants from tobacco smoke were significantly associated with time-to-breastfeeding cessation. Infants fed breast milk for ≥4 months had 42.7% more well-child visits and 50.0% fewer respiratory-related clinic visits, compared to infants fed breast milk <4 months. Data support intervention refinements for mothers from smoking households and making NICU-based healthcare workers aware of risk factors for early breastfeeding cessation.

Diabetes after pregnancy: a study protocol for the derivation and validation of a risk prediction model for 5-year risk of diabetes following pregnancy

BACKGROUND

Pregnancy offers a unique opportunity to identify women at higher future risk of type 2 diabetes mellitus (DM). In pregnancy, a woman has greater engagement with the healthcare system, and certain conditions are more apt to manifest, such as gestational DM (GDM) that are important markers for future DM risk. This study protocol describes the development and validation of a risk prediction model (RPM) for estimating a woman's 5-year risk of developing type 2 DM after pregnancy.

METHODS

Data will be obtained from existing Ontario population-based administrative datasets. The derivation cohort will consist of all women who gave birth in Ontario, Canada between April 2006 and March 2014. Pre-specified predictors will include socio-demographic factors (age at delivery, ethnicity), maternal clinical factors (e.g., body mass index), pregnancy-related events (gestational DM, hypertensive disorders of pregnancy), and newborn factors (birthweight percentile). Incident type 2 DM will be identified by linkage to the Ontario Diabetes Database. Weibull accelerated failure time models will be developed to predict 5-year risk of type 2 DM. Measures of predictive accuracy (Nagelkerke's R2), discrimination (C-statistics), and calibration plots will be generated. Internal validation will be conducted using a bootstrapping approach in 500 samples with replacement, and an optimism-corrected C-statistic will be calculated. External validation of the RPM will be conducted by applying the model in a large population-based pregnancy cohort in Alberta, and estimating the above measures of model performance. The model will be re-calibrated by adjusting baseline hazards and coefficients where appropriate.

DISCUSSION

The derived RPM may help identify women at high risk of developing DM in a 5-year period after pregnancy, thus facilitate lifestyle changes for women at higher risk, as well as more frequent screening for type 2 DM after pregnancy.

Re-evaluation of the comparative effectiveness of bootstrap-based optimism correction methods in the development of multivariable clinical prediction models

Background

Multivariable prediction models are important statistical tools for providing synthetic diagnosis and prognostic algorithms based on patients’ multiple characteristics. Their apparent measures for predictive accuracy usually have overestimation biases (known as ‘optimism’) relative to the actual performances for external populations. Existing statistical evidence and guidelines suggest that three bootstrap-based bias correction methods are preferable in practice, namely Harrell’s bias correction and the .632 and .632+ estimators. Although Harrell’s method has been widely adopted in clinical studies, simulation-based evidence indicates that the .632+ estimator may perform better than the other two methods. However, these methods’ actual comparative effectiveness is still unclear due to limited numerical evidence.

Methods

We conducted extensive simulation studies to compare the effectiveness of these three bootstrapping methods, particularly using various model building strategies: conventional logistic regression, stepwise variable selections, Firth’s penalized likelihood method, ridge, lasso, and elastic-net regression. We generated the simulation data based on the Global Utilization of Streptokinase and Tissue plasminogen activator for Occluded coronary arteries (GUSTO-I) trial Western dataset and considered how event per variable, event fraction, number of candidate predictors, and the regression coefficients of the predictors impacted the performances. The internal validity of C-statistics was evaluated.

Results

Under relatively large sample settings (roughly, events per variable ≥ 10), the three bootstrap-based methods were comparable and performed well. However, all three methods had biases under small sample settings, and the directions and sizes of biases were inconsistent. In general, Harrell’s and .632 methods had overestimation biases when event fraction become lager. Besides, .632+ method had a slight underestimation bias when event fraction was very small. Although the bias of the .632+ estimator was relatively small, its root mean squared error (RMSE) was comparable or sometimes larger than those of the other two methods, especially for the regularized estimation methods.

Conclusions

In general, the three bootstrap estimators were comparable, but the .632+ estimator performed relatively well under small sample settings, except when the regularized estimation methods are adopted.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12874-020-01201-w.

Predicting the first smoking lapse during a quit attempt: A machine learning approach

BACKGROUND

Just-in-time adaptive interventions (JITAI) aim to prevent smoking lapse using tailored support delivered via mobile technology in the moments when it is most needed. Effective smoking cessation JITAI rely on the development of accurate decision rules that determine when someone is most likely to lapse. The primary goal of the present study was to identify the strongest predictors of first lapse among smokers undergoing a quit attempt.

METHODS

Smokers attending a clinic-based smoking cessation program (n = 74) were asked to complete ecological momentary assessments five times daily on study-provided smartphones for 4 weeks post-quit. A three-stage modeling process utilized Cox proportional hazards regression to examine time to lapse a function of 31 predictors. First, univariate models evaluated the relationship between each predictor and time to lapse. Second, the elastic net machine learning algorithm was used to select the best predictors. Third, backwards elimination further reduced the set of predictors to optimize parsimony.

RESULTS

Univariate models identified seven predictors significantly related to time to lapse. The elastic net algorithm retained five: perceived odds of smoking today, confidence in ability to avoid smoking, motivation to avoid smoking, urge to smoke, and cigarette availability. The reduced model demonstrated inadequate approximation to the non-penalized baseline model.

CONCLUSIONS

Accurate estimation of moments of high risk for smoking lapse remains an important goal in the development of JITAI. These results demonstrate the utility of exploratory data-driven approaches to variable selection. The results of this study can inform future JITAI by highlighting targets for intervention.

Sample sizes of prediction model studies in prostate cancer were rarely justified and often insufficient

OBJECTIVE

Developing clinical prediction models (CPMs) on data of sufficient sample size is critical to help minimize overfitting. Using prostate cancer as a clinical exemplar, we aimed to investigate to what extent existing CPMs adhere to recent formal sample size criteria, or historic rules of thumb of events per predictor parameter (EPP)≥10.

STUDY DESIGN AND SETTING

A systematic review to identify CPMs related to prostate cancer, which provided enough information to calculate minimum sample size. We compared the reported sample size of each CPM against the traditional 10 EPP rule of thumb and formal sample size criteria.

RESULTS

About 211 CPMs were included. Three of the studies justified the sample size used, mostly using EPP rules of thumb. Overall, 69% of the CPMs were derived on sample sizes that surpassed the traditional EPP≥10 rule of thumb, but only 48% surpassed recent formal sample size criteria. For most CPMs, the required sample size based on formal criteria was higher than the sample sizes to surpass 10 EPP.

CONCLUSION

Few of the CPMs included in this study justified their sample size, with most justifications being based on EPP. This study shows that, in real-world data sets, adhering to the classic EPP rules of thumb is insufficient to adhere to recent formal sample size criteria.

Development and Validation of HSCAPS-1: A Clinical Decision Support Tool for Diagnosis of Hidradenitis Suppurativa over Cutaneous Abscess

BACKGROUND

A clinical decision support tool may improve recognition of hidradenitis suppurativa (HS) and reduce diagnosis delay.

OBJECTIVE

To develop and initially validate a clinical decision support to predict diagnosis of HS and distinguish it from cutaneous abscess of the axilla, groin, perineum, and buttock.

METHODS

This was a retrospective, cross-sectional analysis between January 2012 and June 2017 (development set) and July 2017 and March 2019 (validation set). We used an electronic records sample of 56 million patients from the Explorys database to identify patients with an ambulatory visit associated with either HS or cutaneous of the axilla, groin, perineum, and buttock. The outcome was predicted probability of HS diagnosis.

RESULTS

Development set included 7,974 patients with mean age of 41.4 years, who were predominantly female (66%) and white (62%). Validation set included 1,560 patients with similar demographic composition. Factors which were stronger independent predictors of HS included female sex (OR 2.17 [95% CI 1.96-2.40]); African American race (1.28 [95% CI 1.15-1.44]); increasing BMI (OR 1.05 [95% CI 1.05-1.06)]; history of acne (OR 3.46 [95% CI 2.83-4.23]); Down syndrome (OR 5.35 [95% CI 2.03-14.12]); and prescription for at least 7 opioid medications in the past year (OR 1.05 [95% CI 0.83-1.33]). Up to age 45 years, increasing age was a stronger predictor of HS diagnosis. The simplified model showed good discrimination (c-statistic 0.746 [SE 0.013]) and moderate calibration (calibration intercept -0.260 [SE 0.055]; calibration slope 1.142 [SE 0.076]).

CONCLUSION

This clinical decision support tool shows good performance in predicting diagnosis of HS and distinguishing it from cutaneous abscess that involves the axilla, groin, perineum, and buttock.

Validation of predictive models identifying patients at risk for massive transfusion during liver transplantation and their potential impact on blood bank resource utilization

BACKGROUND

Intraoperative massive transfusion (MT) is common during liver transplantation (LT). A predictive model of MT has the potential to improve use of blood bank resources.

STUDY DESIGN AND METHODS

Development and validation cohorts were identified among deceased-donor LT recipients from 2010 to 2016. A multivariable model of MT generated from the development cohort was validated with the validation cohort and refined using both cohorts. The combined cohort also validated the previously reported McCluskey risk index (McRI). A simple modified risk index (ModRI) was then created from the combined cohort. Finally, a method to translate model predictions to a population-specific blood allocation strategy was described and demonstrated for the study population.

RESULTS

Of the 403 patients, 60 (29.6%) in the development and 51 (25.5%) in the validation cohort met the definition for MT. The ModRI, derived from variables incorporated into multivariable model, ranged from 0 to 5, where 1 point each was assigned for hemoglobin level of less than 10 g/dL, platelet count of less than 100 × 10⁹ /dL, thromboelastography R interval of more than 6 minutes, simultaneous liver and kidney transplant and retransplantation, and a ModRI of more than 2 defined recipients at risk for MT. The multivariable model, McRI, and ModRI demonstrated good discrimination (c statistic [95% CI], 0.77 [0.70-0.84]; 0.69 [0.62-0.76]; and 0.72 [0.65-0.79], respectively, after correction for optimism). For blood allocation of 6 or 15 units of red blood cells (RBCs) based on risk of MT, the ModRI would prevent unnecessary crossmatching of 300 units of RBCs/100 transplants.

CONCLUSIONS

Risk indices of MT in LT can be effective for risk stratification and reducing unnecessary blood bank resource utilization.

Predictive value of cardiopulmonary fitness parameters in the prognosis of patients with acute coronary syndrome after percutaneous coronary intervention

Objectives

We aimed to determine the predictive value of cardiopulmonary exercise testing (CPX) in the prognosis of patients with acute coronary syndrome (ACS) treated with percutaneous coronary intervention (PCI).

Methods

We conducted a retrospective study including patients who underwent CPX within 1 year of PCI between September 2012 and October 2017. Patients were followed-up until the occurrence of a major adverse cardiac event (MACE) or administrative censoring (September 2019). A Cox regression model was used to identify significant predictors of a MACE. Model performance was evaluated in terms of discrimination (C-statistic) and calibration (calibration-in-the-large).

Results

In total, 184 patients were included and followed-up for a median 51 months (interquartile range: 36–67 months) and 32 events occurred. Multivariable analysis revealed that body mass index and Gensini score were significant predictors of a MACE. Four CPX-related variables were found to be predictive of a MACE: premature CPX termination, peak oxygen uptake, heart rate reserve, and ventilatory equivalent for carbon dioxide slope. The final prediction model had a C-statistic of 0.92 and calibration-in-the-large 0.58%.

Conclusion

CPX-related parameters may have high predictive value for poor outcomes in patients with ACS who undergo PCI, indicating a need for appropriate treatment and timely management.

Preoperative prediction of Bleeding Independently associated with Mortality after noncardiac Surgery (BIMS): an international prospective cohort study

BACKGROUND

Diagnostic criteria for Bleeding Independently associated with Mortality after noncardiac Surgery (BIMS) have been defined as bleeding that leads to a postoperative haemoglobin <70 g L^-1, leads to blood transfusion, or is judged to be the direct cause of death. Preoperative prediction guides for BIMS can facilitate informed consent and planning of perioperative care.

METHODS

In a prospective cohort study of 16 079 participants aged ≥45 yr having inpatient noncardiac surgery at 12 academic hospitals in eight countries between 2007 and 2011, 17.3% (2782) experienced BIMS. An electronic risk calculator for BIMS was developed and internally validated by logistic regression with bootstrapping, and further simplified to a risk index. Decision curve analysis assessed the potential utility of each prediction guide compared with a strategy of identifying risk of BIMS based on preoperative haemoglobin <120 g L^-1.

RESULTS

With information about the type of surgery, preoperative haemoglobin, age, sex, functional status, kidney function, history of high-risk coronary artery disease, and active cancer, the risk calculator accurately predicted BIMS (bias-corrected C-statistic, 0.84; 95% confidence interval, 0.837-0.852). A simplified index based on preoperative haemoglobin <120 g L^-1, open surgery, and high-risk surgery also predicted BIMS, but less accurately (C-statistic, 0.787; 95% confidence interval, 0.779-0.796). Both prediction guides could improve decision making compared with knowledge of haemoglobin <120 g L^-1 alone.

CONCLUSIONS

BIMS, defined as bleeding that leads to a postoperative haemoglobin <70 g L^-1, leads to blood transfusion, or that is judged to be the direct cause of death, can be predicted by a simple risk index before surgery.

CLINICAL TRIAL REGISTRATION

NCT00512109.

Factors Associated with Step Numbers in Acutely Hospitalized Older Adults: The Hospital-Activities of Daily Living Study

OBJECTIVES

To determine the number of steps taken by older patients in hospital and 1 week after discharge; to identify factors associated with step numbers after discharge; and to examine the association between functional decline and step numbers after discharge.

DESIGN

Prospective observational cohort study conducted in 2015-2017.

SETTING AND PARTICIPANTS

Older adults (≥70 years of age) acutely hospitalized for at least 48 hours at internal, cardiology, or geriatric wards in 6 Dutch hospitals.

METHODS

Steps were counted using the Fitbit Flex accelerometer during hospitalization and 1 week after discharge. Demographic, somatic, physical, and psychosocial factors were assessed during hospitalization. Functional decline was determined 1 month after discharge using the Katz activities of daily living index.

RESULTS

The analytic sample included 188 participants [mean age (standard deviation) 79.1 (6.7)]. One month postdischarge, 33 out of 174 participants (19%) experienced functional decline. The median number of steps was 656 [interquartile range (IQR), 250-1146] at the last day of hospitalization. This increased to 1750 (IQR 675-4114) steps 1 day postdischarge, and to 1997 (IQR 938-4098) steps 7 days postdischarge. Age [β = -57.93; 95% confidence interval (CI) -111.15 to -4.71], physical performance (β = 224.95; 95% CI 117.79-332.11), and steps in hospital (β = 0.76; 95% CI 0.46-1.06) were associated with steps postdischarge. There was a significant association between step numbers after discharge and functional decline 1 month after discharge (β = -1400; 95% CI -2380 to -420; P = .005).

CONCLUSIONS AND IMPLICATIONS

Among acutely hospitalized older adults, step numbers double 1 day postdischarge, indicating that their capacity is underutilized during hospitalization. Physical performance and physical activity during hospitalization are key to increasing the number of steps postdischarge. The number of steps 1 week after discharge is a promising indicator of functional decline 1 month after discharge.

Developing and validating a multivariable prediction model for in-hospital mortality of pneumonia with advanced chronic kidney disease patients: a retrospective analysis using a nationwide database in Japan

BACKGROUND

The prognosis of pneumonia in patients with advanced stage chronic kidney disease (CKD) remains unimproved for years. We attempt to develop a simple and more useful scoring system for predicting in-hospital mortality for advanced CKD patients with pneumonia.

METHODS

Using the Diagnosis Procedure Combination database, we identified the in-hospital adult patients both with a record of pneumonia and stage 5 or 5D CKD as a comorbidity on admission between April 1, 2012 and March 31, 2016. Predictive variable selection was analyzed by multivariable logistic regression analysis, stepwise method, LASSO method and random forest method, and then develop a new simple scoring system seeking for highest c-statistics combination of variables in one sample data set for model development. Finally, we compared c-statistics of univariate logistic regression about new scoring system with c-statistics about "A-DROP" in the other sample data set.

RESULT

We identified 8402 patients in 707 hospitals, and the total in-hospital mortality was 11.0% (437 patients) in development data set. Seven variables were selected, which includes age (male ≥ 70 years, female ≥ 75 years), respiratory failure, orientation disturbance, low blood pressure, the need of assistance in feeding or bowel control, severe or moderate thinness and CRP 200 mg/L or extent of consolidation on chest X-ray ≥ 2/3 of one lung. The c-statistics of univariate logistic regression was 0.8017 using seven variables, while that was 0.7372 using "A-DROP" CONCLUSION: In advanced CKD patients, if we select appropriate variables for predicting in-hospital mortality, simple scoring system may have better discrimination than "A-DROP".

Developing a multivariable prediction model for functional outcome after reperfusion therapy for acute ischaemic stroke: study protocol for the Targeting Optimal Thrombolysis Outcomes (TOTO) multicentre cohort study

INTRODUCTION

Intravenous thrombolysis (IVT) with recombinant tissue plasminogen activator (rt-PA) is the only approved pharmacological reperfusion therapy for acute ischaemic stroke. Despite population benefit, IVT is not equally effective in all patients, nor is it without significant risk. Uncertain treatment outcome prediction complicates patient treatment selection. This study will develop and validate predictive algorithms for IVT response, using clinical, radiological and blood-based biomarker measures. A secondary objective is to develop predictive algorithms for endovascular thrombectomy (EVT), which has been proven as an effective reperfusion therapy since study inception.

METHODS AND ANALYSIS

The Targeting Optimal Thrombolysis Outcomes Study is a multicenter prospective cohort study of ischaemic stroke patients treated at participating Australian Stroke Centres with IVT and/or EVT. Patients undergo neuroimaging using multimodal CT or MRI at baseline with repeat neuroimaging 24 hours post-treatment. Baseline and follow-up blood samples are provided for research use. The primary outcome is good functional outcome at 90 days poststroke, defined as a modified Rankin Scale (mRS) Score of 0-2. Secondary outcomes are reperfusion, recanalisation, infarct core growth, change in stroke severity, poor functional outcome, excellent functional outcome and ordinal mRS at 90 days. Primary predictive models will be developed and validated in patients treated only with rt-PA. Models will be built using regression methods and include clinical variables, radiological measures from multimodal neuroimaging and blood-based biomarkers measured by mass spectrometry. Predictive accuracy will be quantified using c-statistics and R². In secondary analyses, models will be developed in patients treated using EVT, with or without prior IVT, reflecting practice changes since original study design.

ETHICS AND DISSEMINATION

Patients, or relatives when patients could not consent, provide written informed consent to participate. This study received approval from the Hunter New England Local Health District Human Research Ethics Committee (reference 14/10/15/4.02). Findings will be disseminated via peer-reviewed publications and conference presentations.

Comparison of the Charlson Comorbidity Index and the Davies Score as a Predictor of Outcomes in PD Patients

Background

Comorbidity is a strong predictor and confounds many studies of outcomes. Previous studies have shown that the Charlson Comorbidity Index (CCI) and the Davies score predict mortality in peritoneal dialysis (PD) patients. However, there are few data on the comparison of comorbidity scores.

Objective

To compare the CCI (combines comorbidity and age) and Davies score (comorbidity score without age) to see if one score was superior to the other in predicting outcomes.

Design

Prospective database study.

Setting

Seven dialysis centers in Western Pennsylvania.

Participants

415 incident PD patients, starting PD from 1/1/90 to 2/1/00.

Measurements

The CCI and Davies score calculated at the start of PD; serum albumin levels and demographics at the start of PD; total hospitalizations and mortality, collected prospectively.

Results

The correlation between CCI and Davies was 0.80, p < 0.0001. The CCI was inversely correlated with serum albumin (–0.31, p < 0.0001). Davies was significantly correlated with age (0.32, p < 0.0001) and inversely correlated with albumin (–0.27, p < 0.0001). The CCI alone was a stronger predictor than Davies alone (score by best subsets regression 49.6 vs 42.0, p = 0.0058). The CCI and Davies with age appeared to be equivalent models of survival (49.61 vs 49.64). The best predictive models were CCI and initial albumin, or Davies, age, and initial albumin. Both CCI and Davies were predictors of hospitalization rates, but the model with the Davies score was better (Akaike information criterion 799.2 vs 850.2). The best predictive model was Davies, albumin, age, and race.

Conclusions

Both comorbidity scores were significant predictors of outcomes, with CCI the stronger predictor for mortality, but the Davies was a stronger predictor of hospitalizations. One or both should be done at the start of dialysis to predict outcome.

Using Elastic Net Penalized Cox Proportional Hazards Regression to Identify Predictors of Imminent Smoking Lapse

Introduction

Machine learning algorithms such as elastic net regression and backward selection provide a unique and powerful approach to model building given a set of psychosocial predictors of smoking lapse measured repeatedly via ecological momentary assessment (EMA). Understanding these predictors may aid in developing interventions for smoking lapse prevention.

Methods

In a randomized-controlled smoking cessation trial, smartphone-based EMAs were collected from 92 participants following a scheduled quit date. This secondary analysis utilized elastic net-penalized cox proportional hazards regression and model approximation via backward elimination to (1) optimize a predictive model of time to first lapse and (2) simplify that model to its core constituent predictors to maximize parsimony and generalizability.

Results

Elastic net proportional hazards regression selected 17 of 26 possible predictors from 2065 EMAs to model time to first lapse. The predictors with the highest magnitude regression coefficients were having consumed alcohol in the past hour, being around and interacting with a smoker, and having cigarettes easily available. This model was reduced using backward elimination, retaining five predictors and approximating to 93.9% of model fit. The retained predictors included those mentioned above as well as feeling irritable and being in areas where smoking is either discouraged or allowed (as opposed to not permitted).

Conclusions

The strongest predictors of smoking lapse were environmental in nature (e.g., being in smoking-permitted areas) as opposed to internal factors such as psychological affect. Interventions may be improved by a renewed focus of interventions on these predictors.

Implications

The present study demonstrated the utility of machine learning algorithms to optimize the prediction of time to smoking lapse using EMA data. The two models generated by the present analysis found that environmental factors were most strongly related to smoking lapse. The results support the use of machine learning algorithms to investigate intensive longitudinal data, and provide a foundation for the development of highly tailored, just-in-time interventions that can target on multiple antecedents of smoking lapse.

Persistent osteoarticular pain in children: early clinical and laboratory findings suggestive of acute lymphoblastic leukemia (a multicenter case-control study of 147 patients)

BACKGROUND

The aim of this study was to identify early clinical and laboratory features that distinguish acute lymphoblastic leukemia (ALL) from juvenile idiopathic arthritis (JIA) in children presenting with persistent bone or joint pain for at least 1 month.

METHODS

We performed a multicenter case-control study and reviewed medical records of children who initially presented with bone or joint pain lasting for at least 1 month, all of whom were given a secondary diagnosis of JIA or ALL, in four French University Hospitals. Each patient with ALL was paired by age with two children with JIA. Logistic regression was used to compare clinical and laboratory data from the two groups.

RESULTS

Forty-nine children with ALL and 98 with JIA were included. The single most important feature distinguishing ALL from JIA was the presence of hepatomegaly, splenomegaly or lymphadenopathy; at least one of these manifestations was present in 37 cases with ALL, but only in 2 controls with JIA, for an odds ratio (OR) of 154 [95%CI: 30-793] (regression coefficient: 5.0). If the presence of these findings is missed or disregarded, multivariate analyses showed that non-articular bone pain and/or general symptoms (asthenia, anorexia or weight loss) (regression coefficient: 4.8, OR 124 [95%CI: 11.4-236]), neutrophils < 2 × 10⁹/L (regression coefficient: 3.9, OR 50 [95%CI: 4.3-58]), and platelets < 300 × 10⁹/L (regression coefficient: 2.6, OR 14 [95%CI: 2.3-83.9]) were associated with the presence of ALL (area under the ROC curve: 0.96 [95%CI: 0.93-0.99]).

CONCLUSIONS

Based on our findings we propose the following preliminary decision tree to be tested in prospective studies: in children presenting with at least 1 month of osteoarticular pain and no obvious ALL in peripheral smear, perform a bone marrow examination if hepatomegaly, splenomegaly or lymphadenopathy is present. If these manifestations are absent, perform a bone marrow examination if there is fever or elevated inflammatory markers associated with non-articular bone pain, general symptoms (asthenia, anorexia or weight loss), neutrophils < 2 × 10⁹/L or platelets < 300 × 10⁹/L.

Development and validation of a cardiovascular disease risk-prediction model using population health surveys: the Cardiovascular Disease Population Risk Tool (CVDPoRT)

BACKGROUND

Routinely collected data from large population health surveys linked to chronic disease outcomes create an opportunity to develop more complex risk-prediction algorithms. We developed a predictive algorithm to estimate 5-year risk of incident cardiovascular disease in the community setting.

METHODS

We derived the Cardiovascular Disease Population Risk Tool (CVDPoRT) using prospectively collected data from Ontario respondents of the Canadian Community Health Surveys, representing 98% of the Ontario population (survey years 2001 to 2007; follow-up from 2001 to 2012) linked to hospital admission and vital statistics databases. Predictors included body mass index, hypertension, diabetes, and multiple behavioural, demographic and general health risk factors. The primary outcome was the first major cardiovascular event resulting in hospital admission or death. Death from a noncardiovascular cause was considered a competing risk.

RESULTS

We included 104 219 respondents aged 20 to 105 years. There were 3709 cardiovascular events and 818 478 person-years follow-up in the combined derivation and validation cohorts (5-year cumulative incidence function, men: 0.026, 95% confidence interval [CI] 0.025-0.028; women: 0.018, 95% 0.017-0.019). The final CVDPoRT algorithm contained 12 variables, was discriminating (men: C statistic 0.82, 95% CI 0.81-0.83; women: 0.86, 95% CI 0.85-0.87) and was well-calibrated in the overall population (5-year observed cumulative incidence function v. predicted risk, men: 0.28%; women: 0.38%) and in nearly all predefined policy-relevant subgroups (206 of 208 groups).

INTERPRETATION

The CVDPoRT algorithm can accurately discriminate cardiovascular disease risk for a wide range of health profiles without the aid of clinical measures. Such algorithms hold potential to support precision medicine for individual or population uses. Study registration: ClinicalTrials.gov, no. NCT02267447.

Development of Prediction Models for Sick Leave Due to Musculoskeletal Disorders

Purpose The aim of this study was to develop prediction models to determine the risk of sick leave due to musculoskeletal disorders (MSD) in non-sick listed employees and to compare models for short-term (i.e., 3 months) and long-term (i.e., 12 months) predictions. Methods Cohort study including 49,158 Dutch employees who participated in occupational health checks between 2009 and 2015 and sick leave data recorded during 12 months follow-up. Prediction models for MSD sick leave within 3 and 12 months after the health check were developed with logistic regression analysis using routinely assessed health check variables. The performance of the prediction models was evaluated with explained variance (Nagelkerke's R-square), calibration (Hosmer-Lemeshow test) and discrimination (area under the receiver operating characteristic curve, AUC) measures. Results A total of 376 (0.8%) and 1193 (2.4%) employees had MSD sick leave within 3 and 12 months after the health check. The prediction models included similar predictor variables (educational level, musculoskeletal complaints, distress, supervisor social support, work-home interference, intrinsic motivation, development opportunities, and work pace). The explained variances were 7.6% and 8.8% for the model with 3 and 12 months follow-up, respectively. Both prediction models showed adequate calibration and discriminated between employees with and without MSD sick leave 3 months (AUC = 0.761; Interquartile range [IQR] 0.759-0.763) and 12 months (AUC = 0.740; IQR 0.738-0.741) after the health check. Conclusion The prediction models could be used to determine the risk of MSD sick leave in non-sick listed employees and invite them to preventive consultations with occupational health providers.

Fundamentals of Clinical Prediction Modeling for the Neurosurgeon

Clinical prediction models in neurosurgery are increasingly reported. These models aim to provide an evidence-based approach to the estimation of the probability of a neurosurgical outcome by combining 2 or more prognostic variables. Model development and model reporting are often suboptimal. A basic understanding of the methodology of clinical prediction modeling is needed when interpreting these models. We address basic statistical background, 7 modeling steps, and requirements of these models such that they may fulfill their potential for major impact for our daily clinical practice and for future scientific work.

Sample size for binary logistic prediction models: Beyond events per variable criteria

Binary logistic regression is one of the most frequently applied statistical approaches for developing clinical prediction models. Developers of such models often rely on an Events Per Variable criterion (EPV), notably EPV ≥10, to determine the minimal sample size required and the maximum number of candidate predictors that can be examined. We present an extensive simulation study in which we studied the influence of EPV, events fraction, number of candidate predictors, the correlations and distributions of candidate predictor variables, area under the ROC curve, and predictor effects on out-of-sample predictive performance of prediction models. The out-of-sample performance (calibration, discrimination and probability prediction error) of developed prediction models was studied before and after regression shrinkage and variable selection. The results indicate that EPV does not have a strong relation with metrics of predictive performance, and is not an appropriate criterion for (binary) prediction model development studies. We show that out-of-sample predictive performance can better be approximated by considering the number of predictors, the total sample size and the events fraction. We propose that the development of new sample size criteria for prediction models should be based on these three parameters, and provide suggestions for improving sample size determination.

External validation of a prognostic model to predict survival of patients with sentinel node-negative melanoma

Background

Identifying patients with sentinel node‐negative melanoma at high risk of recurrence or death is important. The European Organisation for Research and Treatment of Cancer (EORTC) recently developed a prognostic model including Breslow thickness, ulceration and site of the primary tumour. The aims of the present study were to validate this prognostic model externally and to assess whether it could be improved by adding other prognostic factors.

Methods

Patients with sentinel node‐negative cutaneous melanoma were included in this retrospective single‐institution study. The β values of the EORTC prognostic model were used to predict recurrence‐free survival and melanoma‐specific survival. The predictive performance was assessed by discrimination (c‐index) and calibration. Seeking to improve the performance of the model, additional variables were added to a Cox proportional hazards model.

Results

Some 4235 patients with sentinel node‐negative cutaneous melanoma were included. The median follow‐up time was 50 (i.q.r. 18·5–81·5) months. Recurrences and deaths from melanoma numbered 793 (18·7 per cent) and 456 (10·8 per cent) respectively. Validation of the EORTC model showed good calibration for both outcomes, and a c‐index of 0·69. The c‐index was only marginally improved to 0·71 when other significant prognostic factors (sex, age, tumour type, mitotic rate) were added.

Conclusion

This study validated the EORTC prognostic model for recurrence‐free and melanoma‐specific survival of patients with negative sentinel nodes. The addition of other prognostic factors only improved the model marginally. The validated EORTC model could be used for personalizing follow‐up and selecting high‐risk patients for trials of adjuvant systemic therapy.

Predicting Mobility Limitations in Patients With Total Knee Arthroplasty in the Inpatient Setting

OBJECTIVE

To develop a prediction model for postoperative day 3 mobility limitations in patients undergoing total knee arthroplasty (TKA).

DESIGN

Prospective cohort study.

SETTING

Inpatients in a tertiary care hospital.

PARTICIPANTS

A sample of patients (N=2300) who underwent primary TKA in 2016-2017.

INTERVENTIONS

Not applicable.

MAIN OUTCOME MEASURE

Candidate predictors included demographic variables and preoperative clinical and psychosocial measures. The outcome of interest was mobility limitations on post-TKA day 3, and this was determined a priori by an ordinal mobility outcome hierarchy based on the type of the gait aids prescribed and the level of physiotherapist assistance provided. To develop the model, we fitted a multivariable proportional odds regression model with bootstrap internal validation. We used a model approximation approach to create a simplified model that approximated predictions from the full model with 95% accuracy.

RESULTS

On post-TKA day 3, 11% of patients required both walkers and therapist assistance to ambulate safely. Our prediction model had a concordance index of 0.72 (95% confidence interval, 0.68-0.75) when evaluating these patients. In the simplified model, predictors of greater mobility limitations included older age, greater walking aid support required preoperatively, less preoperative knee flexion range of movement, low-volume surgeon, contralateral knee pain, higher body mass index, non-Chinese race, and greater self-reported walking limitations preoperatively.

CONCLUSION

We have developed a prediction model to identify patients who are at risk for mobility limitations in the inpatient setting. When used preoperatively as part of a shared-decision making process, it can potentially influence rehabilitation strategies and facilitate discharge planning.

Sample size considerations and predictive performance of multinomial logistic prediction models

Multinomial Logistic Regression (MLR) has been advocated for developing clinical prediction models that distinguish between three or more unordered outcomes. We present a full‐factorial simulation study to examine the predictive performance of MLR models in relation to the relative size of outcome categories, number of predictors and the number of events per variable. It is shown that MLR estimated by Maximum Likelihood yields overfitted prediction models in small to medium sized data. In most cases, the calibration and overall predictive performance of the multinomial prediction model is improved by using penalized MLR. Our simulation study also highlights the importance of events per variable in the multinomial context as well as the total sample size. As expected, our study demonstrates the need for optimism correction of the predictive performance measures when developing the multinomial logistic prediction model. We recommend the use of penalized MLR when prediction models are developed in small data sets or in medium sized data sets with a small total sample size (ie, when the sizes of the outcome categories are balanced). Finally, we present a case study in which we illustrate the development and validation of penalized and unpenalized multinomial prediction models for predicting malignancy of ovarian cancer.

Obstructive sleep apnea in patients surviving acute hypercapnic respiratory failure is best predicted by static hyperinflation

Rationale

Acute hypercapnic respiratory failure (AHRF) treated with non-invasive ventilation in the ICU is frequently caused by chronic obstructive pulmonary disease (COPD) exacerbations and obesity-hypoventilation syndrome, the latter being most often associated with obstructive sleep apnea. Overlap syndrome (a combination of COPD and obstructive sleep apnea) may represent a major burden in this population, and specific diagnostic pathways are needed to improve its detection early after ICU discharge.

Objectives

To evaluate whether pulmonary function tests can identify a high probability of obstructive sleep apnea in AHRF survivors and outperform common screening questionnaires to identify the disorder.

Methods

Fifty-three patients surviving AHRF (31 males; median age 67 years (interquartile range: 62–74) participated in the study. Anthropometric data were recorded and body plethysmography was performed 15 days after ICU discharge. A sleep study was performed 3 months after ICU discharge.

Results

The apnea-hypopnea index was negatively associated with static hyperinflation as measured by the residual volume to total lung capacity ratio in the % of predicted (coefficient = -0.64; standard error 0.17; 95% CI -0.97 to -0.31; p<0.001). A similar association was observed in COPD patients only: coefficient = -0.65; standard error 0.19; 95% CI -1.03 to -0.26; p = 0.002. Multivariate analysis with penalized maximum likelihood confirmed that the residual volume to total lung capacity ratio was the main contributor for apnea-hypopnea index variance in addition to classic predictors. Screening questionnaires to select patients at risk for sleep-disordered breathing did not perform well.

Conclusions

In AHRF survivors, static hyperinflation is negatively associated with the apnea-hypopnea index in both COPD and non-COPD patients. Measuring static hyperinflation in addition to classic predictors may help to increase the recognition of obstructive sleep apnea as common screening tools are of limited value in this specific population.

Genetic and Psychosocial Predictors of Aggression: Variable Selection and Model Building With Component-Wise Gradient Boosting

Rationale: Given datasets with a large or diverse set of predictors of aggression, machine learning (ML) provides efficient tools for identifying the most salient variables and building a parsimonious statistical model. ML techniques permit efficient exploration of data, have not been widely used in aggression research, and may have utility for those seeking prediction of aggressive behavior.

Objectives: The present study examined predictors of aggression and constructed an optimized model using ML techniques. Predictors were derived from a dataset that included demographic, psychometric and genetic predictors, specifically FK506 binding protein 5 (FKBP5) polymorphisms, which have been shown to alter response to threatening stimuli, but have not been tested as predictors of aggressive behavior in adults.

Methods: The data analysis approach utilized component-wise gradient boosting and model reduction via backward elimination to: (a) select variables from an initial set of 20 to build a model of trait aggression; and then (b) reduce that model to maximize parsimony and generalizability.

Results: From a dataset of N = 47 participants, component-wise gradient boosting selected 8 of 20 possible predictors to model Buss-Perry Aggression Questionnaire (BPAQ) total score, with R² = 0.66. This model was simplified using backward elimination, retaining six predictors: smoking status, psychopathy (interpersonal manipulation and callous affect), childhood trauma (physical abuse and neglect), and the FKBP5_13 gene (rs1360780). The six-factor model approximated the initial eight-factor model at 99.4% of R².

Conclusions: Using an inductive data science approach, the gradient boosting model identified predictors consistent with previous experimental work in aggression; specifically psychopathy and trauma exposure. Additionally, allelic variants in FKBP5 were identified for the first time, but the relatively small sample size limits generality of results and calls for replication. This approach provides utility for the prediction of aggression behavior, particularly in the context of large multivariate datasets.

Dementia Population Risk Tool (DemPoRT): study protocol for a predictive algorithm assessing dementia risk in the community

INTRODUCTION

The burden of disease from dementia is a growing global concern as incidence increases dramatically with age, and average life expectancy has been increasing around the world. Planning for an ageing population requires reliable projections of dementia prevalence; however, existing population projections are simple and have poor predictive accuracy. The Dementia Population Risk Tool (DemPoRT) will predict incidence of dementia in the population setting using multivariable modelling techniques and will be used to project dementia prevalence.

METHODS AND ANALYSIS

The derivation cohort will consist of elderly Ontario respondents of the Canadian Community Health Survey (CCHS) (2001, 2003, 2005 and 2007; 18 764 males and 25 288 females). Prespecified predictors include sociodemographic, general health, behavioural, functional and health condition variables. Incident dementia will be identified through individual linkage of survey respondents to population-level administrative healthcare databases (1797 and 3281 events, and 117 795 and 166 573 person-years of follow-up, for males and females, respectively, until 31 March 2014). Using time of first dementia capture as the primary outcome and death as a competing risk, sex-specific proportional hazards regression models will be estimated. The 2008/2009 CCHS survey will be used for validation (approximately 4600 males and 6300 females). Overall calibration and discrimination will be assessed as well as calibration within predefined subgroups of importance to clinicians and policy makers.

ETHICS AND DISSEMINATION

Research ethics approval has been granted by the Ottawa Health Science Network Research Ethics Board. DemPoRT results will be submitted for publication in peer-review journals and presented at scientific meetings. The algorithm will be assessable online for both population and individual uses.

TRIAL REGISTRATION NUMBER

ClinicalTrials.gov NCT03155815, pre-results.

Absence of ultrasound inflammation in patients presenting with arthralgia rules out the development of arthritis

BACKGROUND

To decrease the burden of disease of rheumatoid arthritis (RA), patients at risk for RA need to be identified as early as possible, preferably when no clinically apparent synovitis can be detected. Up to now, it has been fairly difficult to identify those patients with arthralgia who develop inflammatory arthritis (IA), but recent studies using ultrasound (US) suggest that earlier detection is possible. We aimed to identify patients with arthralgia developing IA within 1 year using US to detect subclinical synovitis at first consultation.

METHODS

In a multi-centre cohort study, we followed patients with arthralgia with at least two painful joints of the hands, feet or shoulders without clinical synovitis over 1 year. Symptom duration was < 1 year, and symptoms were not explained by other conditions. At baseline and at 6 and 12 months, data were collected for physical examinations, laboratory values and diagnoses. At baseline, we examined 26 joints ultrasonographically (bilateral metacarpophalangeal joints 2-5, proximal interphalangeal joints 2-5, wrist and metatarsophalangeal joints 2-5). Scoring was done semi-quantitatively on greyscale (GS; 0-3) and power Doppler (PD; 0-3) images. US synovitis was defined as GS ≥ 2 and/or PD ≥ 1. IA was defined as clinical soft tissue swelling. Sensitivity and specificity were used to assess the diagnostic value of US for the development of IA. Univariate logistic regression was used to analyse the association between independent variables and the incidence of IA. For multivariate logistic regression, the strongest variables (p < 0.157) were selected. Missing values for independent variables were imputed.

RESULTS

A total of 196 patients were included, and 159 completed 12 months of follow-up. Thirty-one (16%) patients developed IA, of whom 59% showed US synovitis at baseline. The sensitivity and specificity of US synovitis were 59% and 68%, respectively. If no joints were positive on US, negative predictive value was 89%. In the multivariate logistic regression, age (OR 1.1), the presence of morning stiffness for > 30 minutes (OR 3.3) and PD signal (OR 3.4) were associated with incident IA.

CONCLUSIONS

The presence of PD signal, morning stiffness for > 30 minutes and age at baseline were independently associated with the development of IA. Regarding the value of US in the diagnostic workup of patients with early arthralgia at risk for IA, US did perform well in ruling out IA in patients who did not have US synovitis.

Bleeding impacting mortality after noncardiac surgery: a protocol to establish diagnostic criteria, estimate prognostic importance, and develop and validate a prediction guide in an international prospective cohort study

INTRODUCTION

Various definitions of bleeding have been used in perioperative studies without systematic assessment of the diagnostic criteria for their independent association with outcomes important to patients. Our proposed definition of bleeding impacting mortality after noncardiac surgery (BIMS) is bleeding that is independently associated with death during or within 30 days after noncardiac surgery. We describe our analysis plan to sequentially 1) establish the diagnostic criteria for BIMS, 2) estimate the independent contribution of BIMS to 30-day mortality and 3) develop and internally validate a clinical prediction guide to estimate patient-specific risk of BIMS.

METHODS

In the Vascular Events In Noncardiac Surgery Patients Cohort Evaluation (VISION) study, we prospectively collected bleeding data for 16 079 patients aged 45 years or more who had noncardiac inpatient surgery between 2007 and 2011 at 12 centres in 8 countries across 5 continents. We will include bleeding features independently associated with 30-day mortality in the diagnostic criteria for BIMS. Candidate features will include the need for reoperation due to bleeding, the number of units of erythrocytes transfused, the lowest postoperative hemoglobin concentration, and the absolute and relative decrements in hemoglobin concentration from the preoperative value. We will then estimate the incidence of BIMS and its independent association with 30-day mortality. Last, we will construct and internally validate a clinical prediction guide for BIMS.

INTERPRETATION

This study will address an important gap in our knowledge about perioperative bleeding, with implications for the 200 million patients who undergo noncardiac surgery globally every year. Trial registration: ClinicalTrials.gov, no NCT00512109.

Initial assessment of the infant with neonatal cholestasis-Is this biliary atresia?

Introduction

Optimizing outcome in biliary atresia (BA) requires timely diagnosis. Cholestasis is a presenting feature of BA, as well as other diagnoses (Non-BA). Identification of clinical features of neonatal cholestasis that would expedite decisions to pursue subsequent invasive testing to correctly diagnose or exclude BA would enhance outcomes. The analytical goal was to develop a predictive model for BA using data available at initial presentation.

Methods

Infants at presentation with neonatal cholestasis (direct/conjugated bilirubin >2 mg/dl [34.2 μM]) were enrolled prior to surgical exploration in a prospective observational multi-centered study (PROBE–NCT00061828). Clinical features (physical findings, laboratory results, gallbladder sonography) at enrollment were analyzed. Initially, 19 features were selected as candidate predictors. Two approaches were used to build models for diagnosis prediction: a hierarchical classification and regression decision tree (CART) and a logistic regression model using a stepwise selection strategy.

Results

In PROBE April 2004-February 2014, 401 infants met criteria for BA and 259 for Non-BA. Univariate analysis identified 13 features that were significantly different between BA and Non-BA. Using a CART predictive model of BA versus Non-BA (significant factors: gamma-glutamyl transpeptidase, acholic stools, weight), the receiver operating characteristic area under the curve (ROC AUC) was 0.83. Twelve percent of BA infants were misclassified as Non-BA; 17% of Non-BA infants were misclassified as BA. Stepwise logistic regression identified seven factors in a predictive model (ROC AUC 0.89). Using this model, a predicted probability of >0.8 (n = 357) yielded an 81% true positive rate for BA; <0.2 (n = 120) yielded an 11% false negative rate.

Conclusion

Despite the relatively good accuracy of our optimized prediction models, the high precision required for differentiating BA from Non-BA was not achieved. Accurate identification of BA in infants with neonatal cholestasis requires further evaluation, and BA should not be excluded based only on presenting clinical features.

Algorithm for predicting death among older adults in the home care setting: study protocol for the Risk Evaluation for Support: Predictions for Elder-life in the Community Tool (RESPECT)

INTRODUCTION

Older adults living in the community often have multiple, chronic conditions and functional impairments. A challenge for healthcare providers working in the community is the lack of a predictive tool that can be applied to the broad spectrum of mortality risks observed and may be used to inform care planning.

OBJECTIVE

To predict survival time for older adults in the home care setting. The final mortality risk algorithm will be implemented as a web-based calculator that can be used by older adults needing care and by their caregivers.

DESIGN

Open cohort study using the Resident Assessment Instrument for Home Care (RAI-HC) data in Ontario, Canada, from 1 January 2007 to 31 December 2013.

PARTICIPANTS

The derivation cohort will consist of ∼437 000 older adults who had an RAI-HC assessment between 1 January 2007 and 31 December 2012. A split sample validation cohort will include ∼122 000 older adults with an RAI-HC assessment between 1 January and 31 December 2013.

MAIN OUTCOME MEASURES

Predicted survival from the time of an RAI-HC assessment. All deaths (n≈245 000) will be ascertained through linkage to a population-based registry that is maintained by the Ministry of Health in Ontario.

STATISTICAL ANALYSIS

Proportional hazards regression will be estimated after assessment of assumptions. Predictors will include sociodemographic factors, social support, health conditions, functional status, cognition, symptoms of decline and prior healthcare use. Model performance will be evaluated for 6-month and 12-month predicted risks, including measures of calibration (eg, calibration plots) and discrimination (eg, c-statistics). The final algorithm will use combined development and validation data.

ETHICS AND DISSEMINATION

Research ethics approval has been granted by the Sunnybrook Health Sciences Centre Review Board. Findings will be disseminated through presentations at conferences and in peer-reviewed journals.

TRIAL REGISTRATION NUMBER

NCT02779309, Pre-results.