The implications of propensity score variable selection strategies in pharmacoepidemiology: an empirical illustration

Prognostic score-based model averaging approach for propensity score estimation

BACKGROUND

Propensity scores (PS) are typically evaluated using balance metrics that focus on covariate balance, often without considering their predictive power for the outcome. This approach may not always result in optimal bias reduction in the treatment effect estimate. To address this issue, evaluating covariate balance through prognostic scores, which account for the relationship between covariates and the outcome, has been proposed. Similarly, using a typical model averaging approach for PS estimation that minimizes prediction error for treatment status and covariate imbalance does not necessarily optimize PS-based confounding adjustment. As an alternative approach, using the averaged PS model that minimizes inter-group differences in the prognostic score may further reduce bias in the treatment effect estimate. Moreover, since the prognostic score is also an estimated quantity, model averaging in the prognostic scores can help identify a better prognostic score model. Utilizing the model-averaged prognostic scores as the balance metric for constructing the averaged PS model can contribute to further decreasing bias in treatment effect estimates. This paper demonstrates the effectiveness of the PS model averaging approach based on prognostic score balance and proposes a method that uses the model-averaged prognostic score as a balance metric, evaluating its performance through simulations and empirical analysis.

METHODS

We conduct a series of simulations alongside an analysis of empirical observational data to compare the performances of weighted treatment effect estimates using the proposed and existing approaches. In our examination, we separately provid four candidate estimates for the PS and prognostic score models using traditional regression and machine learning methods. The model averaging of PS based on these candidate estimators is performed to either maximize the prediction accuracy of the treatment or to minimize intergroup differences in covariate distributions or prognostic scores. We also utilize not only the prognostic scores from each candidate model but also an averaged score that best predicted the outcome, for the balance assessment.

RESULTS

The simulation and empirical data analysis reveal that our proposed model-averaging approaches for PS estimation consistently yield lower bias and less variability in treatment effect estimates across various scenarios compared to existing methods. Specifically, using the optimally averaged prognostic scores as a balance metric significantly improves the robustness of the weighted treatment effect estimates.

DISCUSSION

The prognostic score-based model averaging approach for estimating PS can outperform existing model averaging methods. In particular, the estimator using the model averaging prognostic score as a balance metric can produce more robust estimates. Since our results are obtained under relatively simple conditions, applying them to real data analysis requires adjustments to obtain accurate estimates according to the complexity and dimensionality of the data.

CONCLUSIONS

Using the prognostic score as the balance metric for the PS model averaging enhances the performance of the treatment effect estimator, which can be recommended for a wide variety of situations. When applying the proposed method to real-world data, it is important to use it in conjunction with techniques that mitigate issues arising from the complexity and high dimensionality of the data.

Intravenous antiplatelet therapy in patients with ST-segment elevation myocardial infarction undergoing primary percutaneous coronary intervention : A report from the INVEST-STEMI group

The use of intravenous antiplatelet therapy during primary percutaneous coronary intervention (PPCI) is not fully standardized. The aim is to evaluate the effectiveness and safety of periprocedural intravenous administration of cangrelor or tirofiban in a contemporary ST-segment elevation myocardial infarction (STEMI) population undergoing PPCI. This was a multicenter prospective cohort study including consecutive STEMI patients who received cangrelor or tirofiban during PPCI at seven Italian centers. The primary effectiveness measure was the angiographic evidence of thrombolysis in myocardial infarction (TIMI) flow < 3 after PPCI. The primary safety outcome was the in-hospital occurrence of BARC (Bleeding Academic Research Consortium) 2-5 bleedings. The study included 627 patients (median age 63 years, 79% males): 312 received cangrelor, 315 tirofiban. The percentage of history of bleeding, pulmonary edema and cardiogenic shock at admission was comparable between groups. Patients receiving cangrelor had lower ischemia time compared to tirofiban. TIMI flow before PPCI and TIMI thrombus grade were comparable between groups. At propensity score-weighted regression analysis, the risk of TIMI flow < 3 was significantly lower in patients treated with cangrelor compared to tirofiban (adjusted OR: 0.40; 95% CI: 0.30-0.53). The risk of BARC 2-5 bleeding was comparable between groups (adjusted OR:1.35; 95% CI: 0.92-1.98). These results were consistent across multiple prespecified subgroups, including subjects stratified for different total ischemia time, with no statistical interaction. In this real-world multicenter STEMI population, the use of cangrelor was associated with improved myocardial perfusion assessed by coronary angiography after PPCI without increasing clinically-relevant bleedings compared to tirofiban.

High-Loading-Dose Colistin with Nebulized Administration for Carbapenem-Resistant Acinetobacter baumannii Pneumonia in Critically Ill Patients: A Retrospective Cohort Study

Carbapenem-resistant Acinetobacter baumannii (CRAB) infections pose a serious threat, with high morbidity and mortality rates. This retrospective cohort study, conducted at Nakornping Hospital between January 2015 and October 2022, aimed to evaluate the efficacy and safety of a high loading dose (LD) of colistin combined with nebulized colistin in critically ill patients with CRAB pneumonia. Of the 261 patients included, 95 received LD colistin, and 166 received LD colistin with nebulized colistin. Multivariate Cox regression analysis, adjusted for baseline covariates using inverse probability weighting, showed no significant difference in 30-day survival between patients who received LD colistin and those who received LD colistin with nebulized colistin (adjusted hazard ratio [aHR]: 1.17, 95% confidence interval [CI]: 0.80-1.72, p = 0.418). Likewise, there were no significant differences in clinical response (aHR: 0.93, 95% CI: 0.66-1.31, p = 0.688), microbiological response (aHR: 1.21, 95% CI: 0.85-1.73, p = 0.279), or nephrotoxicity (aHR: 1.14, 95% CI: 0.79-1.64, p = 0.492) between the two treatment groups. No significant adverse events related to nebulized colistin were reported. These findings suggest that the addition of nebulized colistin may not offer additional benefits in terms of 30-day survival, clinical or microbiological response, or nephrotoxicity in these patients.

Adaptability of High Dimensional Propensity Score Procedure in the Transition from ICD-9 to ICD-10 in the US Healthcare System

Background

High-Dimensional Propensity Score procedure (HDPS) is a data-driven approach to assist control for confounding in pharmacoepidemiologic research. The transition to the International Classification of Disease (ICD-9/10) in the US health system may pose uncertainty in applying the HDPS procedure.

Methods

We assembled a base cohort of patients in MarketScan^® Commercial Claims Database who had newly initiated celecoxib or traditional NSAIDs to compare gastrointestinal bleeding risk. We then created bootstrapped hypothetical cohorts from the base cohort with predefined patient selection patterns from the ICD eras. Three strategies for HDPS deployment were tested: 1) split the cohort by ICD era, deploy HDPS twice, and pool the relative risks (pooled RR), 2) consider codes from each ICD era as a separate data dimension and deploy HDPS in the entire cohort (data dimensions) and 3) map ICD codes from both eras to Clinical Classifications Software (CCS) concepts before deploying HDPS in the entire cohort (CCS mapping). We calculated percent bias and root-mean-squared error to compare the strategies.

Results

A similar bias reduction was observed in cohorts where patient selection pattern from each ICD era was comparable between the exposure groups. In the presence of considerable disparity in patient selection, we observed a bimodal distribution of propensity scores in the data dimensions strategy, indicating instrument-like covariates. Moreover, the CCS mapping strategy resulted in at least 30% less bias than pooled RR and data dimensions strategies (RMSE: 0.14, 0.19, 0.21, respectively) in this scenario.

Conclusion

Mapping ICD codes to a stable terminology like CCS serves as a helpful strategy to reduce residual bias when deploying HDPS in pharmacoepidemiologic studies spanning both ICD eras.

High-dimensional propensity scores for empirical covariate selection in secondary database studies: Planning, implementation, and reporting

Real-world evidence used for regulatory, payer, and clinical decision-making requires principled epidemiology in design and analysis, applying methods to minimize confounding given the lack of randomization. One technique to deal with potential confounding is propensity score (PS) analysis, which allows for the adjustment for measured preexposure covariates. Since its first publication in 2009, the high-dimensional propensity score (hdPS) method has emerged as an approach that extends traditional PS covariate selection to include large numbers of covariates that may reduce confounding bias in the analysis of healthcare databases. hdPS is an automated, data-driven analytic approach for covariate selection that empirically identifies preexposure variables and proxies to include in the PS model. This article provides an overview of the hdPS approach and recommendations on the planning, implementation, and reporting of hdPS used for causal treatment-effect estimations in longitudinal healthcare databases. We supply a checklist with key considerations as a supportive decision tool to aid investigators in the implementation and transparent reporting of hdPS techniques, and to aid decision-makers unfamiliar with hdPS in the understanding and interpretation of studies employing this approach. This article is endorsed by the International Society for Pharmacoepidemiology.

Applications of propensity score matching: a case series of articles published in Annals of Coloproctology

Propensity score matching (PSM) is an increasingly applied method of ensuring comparability between groups of interest. However, PSM is often applied unconditionally, without precise considerations. The purpose of this study is to provide a nonmathematical guide for clinicians at the stage of designing a PSM-based study. We provide a seed of thought for considering whether applying PSM would be appropriate and, if so, the scope of the list of variables. Although PSM may be simple, its results could vary substantially according to how the propensity score is constructed. Misleading results can be avoided through a critical review of the process of PSM.

Effectiveness and Nephrotoxicity of Loading Dose Colistin-Meropenem versus Loading Dose Colistin-Imipenem in the Treatment of Carbapenem-Resistant Acinetobacter baumannii Infection

Carbapenem-resistant Acinetobacter baumannii (CRAB) is becoming more widely recognized as a serious cause of nosocomial infections, and colistin has been reintroduced in recent years for the treatment of CRAB infection. Combinations of colistin and meropenem or imipenem have been found to be effective against CRAB isolates, whereas clinical investigations have not definitively demonstrated the theoretical benefits of colistin combined therapy in patients with CRAB infections. The objective of this study was to compare the primary outcome (30-day survival rate) and secondary outcomes (clinical response, microbiological response and nephrotoxicity) between patients who received loading dose (LD) colistin−meropenem and LD colistin−imipenem for the treatment of CRAB infection. A retrospective cohort analysis was performed at Chiang Mai University Hospital in patients with CRAB infection who received LD colistin−meropenem or LD colistin−imipenem between 2011 and 2017, and 379 patients fulfilled the requirements for the inclusion criteria. The results of this study showed that patients who received LD colistin−imipenem had a lower 30-day survival rate (adjusted HR = 0.57, 95% CI: 0.37−0.90; p = 0.015) and a lower clinical response (aHR = 0.56, 95% CI: 0.35−0.90; p = 0.017) compared with those who received LD colistin−meropenem. The microbiological response in patients with LD colistin−imipenem was 0.52 times (aHR) lower than that in those who received colistin−meropenem (95% CI: 0.34−0.81; p = 0.004); however, there was no significant difference in nephrotoxicity (aHR = 1.03, 95% CI: 0.67−1.57; p = 0.897) between the two combination regimens. In conclusion, when comparing the combination of LD colistin with imipenem or meropenem, the combination of LD colistin and meropenem provides a better survival rate for treating CRAB. Thus, we suggest that combinations of LD colistin and meropenem should be considered when treating CRAB infections.

Bias amplification in the g-computation algorithm for time-varying treatments: a case study of industry payments and prescription of opioid products

Background

It is often challenging to determine which variables need to be included in the g-computation algorithm under the time-varying setting. Conditioning on instrumental variables (IVs) is known to introduce greater bias when there is unmeasured confounding in the point-treatment settings, and this is also true for near-IVs which are weakly associated with the outcome not through the treatment. However, it is unknown whether adjusting for (near-)IVs amplifies bias in the g-computation algorithm estimators for time-varying treatments compared to the estimators ignoring such variables. We thus aimed to compare the magnitude of bias by adjusting for (near-)IVs across their different relationships with treatments in the time-varying settings.

Methods

After showing a case study of the association between the receipt of industry payments and physicians’ opioid prescribing rate in the US, we demonstrated Monte Carlo simulation to investigate the extent to which the bias due to unmeasured confounders is amplified by adjusting for (near-)IV across several g-computation algorithms.

Results

In our simulation study, adjusting for a perfect IV of time-varying treatments in the g-computation algorithm increased bias due to unmeasured confounding, particularly when the IV had a strong relationship with the treatment. We also found the increase in bias even adjusting for near-IV when such variable had a very weak association with unmeasured confounders between the treatment and the outcome compared to its association with the time-varying treatments. Instead, this bias amplifying feature was not observed (i.e., bias due to unmeasured confounders decreased) by adjusting for near-IV when it had a stronger association with the unmeasured confounders (≥0.1 correlation coefficient in our multivariate normal setting).

Conclusion

It would be recommended to avoid adjusting for perfect IV in the g-computation algorithm to obtain a less biased estimate of the time-varying treatment effect. On the other hand, it may be recommended to include near-IV in the algorithm unless their association with unmeasured confounders is very weak. These findings would help researchers to consider the magnitude of bias when adjusting for (near-)IVs and select variables in the g-computation algorithm for the time-varying setting when they are aware of the presence of unmeasured confounding.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12874-022-01563-3.

Birth Outcomes in Women Who Have Taken Hydroxycholoroquine During Pregnancy: A Prospective Cohort Study

OBJECTIVE

Findings from previous small studies have been reassuring regarding the safety of treatment with hydroxychloroquine (HCQ) during pregnancy. In one recent study, it was demonstrated that the frequency of major birth defects was increased in women who had received HCQ at a dose of ≥400 mg/day during pregnancy. This study was undertaken to examine pregnancy outcomes among women following the use of HCQ.

METHODS

The study cohort comprised pregnant women who were prospectively enrolled in the MotherToBaby/Organization of Teratology Information Specialists Autoimmune Diseases in Pregnancy Study and were receiving treatment with HCQ. For the control groups, disease-matched women without HCQ exposure and healthy women were randomly selected from the same source, with subject matching using a 1:1 ratio. Data were collected through interviews, medical records, and dysmorphology examinations. Pregnancy outcome measures included the presence or absence of major and minor birth defects, rates of spontaneous abortion, rates of preterm delivery, and infant growth measures.

RESULTS

Between 2004 and 2018, 837 pregnant women met the criteria for study inclusion, including 279 women exposed to HCQ during pregnancy and 279 women in each unexposed control group. Sixty pregnant women (7.2%) were lost to follow-up. Among the women with live births, major birth defects occurred as a pregnancy outcome in 20 (8.6%) of 232 women with HCQ exposure in the first trimester, compared to 19 (7.4%) of 256 disease-matched unexposed controls (odds ratio [OR] 1.18, 95% confidence interval [95% CI] 0.61-2.26) and 13 (5.4%) of 239 healthy controls (adjusted OR 0.76, 95% CI 0.28-2.05). Risks did not differ in women who were receiving an HCQ dose of ≥400 mg/day. No pattern of birth defects was identified. There were no differences in the rates of spontaneous abortion or preterm delivery between groups. Occurrence of infant growth deficiencies did not differ in the HCQ-exposed group compared to the disease-matched unexposed control group, except in the infant's head circumference at birth (adjusted OR 1.85, 95% CI 1.07-3.20).

CONCLUSION

In this study, there was no evidence of an increased risk of structural birth defects or other adverse outcomes among women receiving HCQ during pregnancy, with the exception of infant head circumference at birth. For pregnant women being treated with HCQ, these findings are reassuring.

Variable selection in Propensity Score Adjustment to mitigate selection bias in online surveys

The development of new survey data collection methods such as online surveys has been particularly advantageous for social studies in terms of reduced costs, immediacy and enhanced questionnaire possibilities. However, many such methods are strongly affected by selection bias, leading to unreliable estimates. Calibration and Propensity Score Adjustment (PSA) have been proposed as methods to remove selection bias in online nonprobability surveys. Calibration requires population totals to be known for the auxiliary variables used in the procedure, while PSA estimates the volunteering propensity of an individual using predictive modelling. The variables included in these models must be carefully selected in order to maximise the accuracy of the final estimates. This study presents an application, using synthetic and real data, of variable selection techniques developed for knowledge discovery in data to choose the best subset of variables for propensity estimation. We also compare the performance of PSA using different classification algorithms, after which calibration is applied. We also present an application of this methodology in a real-world situation, using it to obtain estimates of population parameters. The results obtained show that variable selection using appropriate methods can provide less biased and more efficient estimates than using all available covariates.

Transparency of high-dimensional propensity score analyses: guidance for diagnostics and reporting

Purpose

The high‐dimensional propensity score (HDPS) is a semi‐automated procedure for confounder identification, prioritisation and adjustment in large healthcare databases that requires investigators to specify data dimensions, prioritisation strategy and tuning parameters. In practice, reporting of these decisions is inconsistent and this can undermine the transparency, and reproducibility of results obtained. We illustrate reporting tools, graphical displays and sensitivity analyses to increase transparency and facilitate evaluation of the robustness of analyses involving HDPS.

Methods

Using a study from the UK Clinical Practice Research Datalink that implemented HDPS we demonstrate the application of the proposed recommendations.

Results

We identify seven considerations surrounding the implementation of HDPS, such as the identification of data dimensions, method for code prioritisation and number of variables selected. Graphical diagnostic tools include assessing the balance of key confounders before and after adjusting for empirically selected HDPS covariates and the identification of potentially influential covariates. Sensitivity analyses include varying the number of covariates selected and assessing the impact of covariates behaving empirically as instrumental variables. In our example, results were robust to both the number of covariates selected and the inclusion of potentially influential covariates. Furthermore, our HDPS models achieved good balance in key confounders.

Conclusions

The data‐adaptive approach of HDPS and the resulting benefits have led to its popularity as a method for confounder adjustment in pharmacoepidemiological studies. Reporting of HDPS analyses in practice may be improved by the considerations and tools proposed here to increase the transparency and reproducibility of study results.

Degree of Safety Against Falls Provided by 4 Different Prosthetic Knee Types in People With Transfemoral Amputation: A Retrospective Observational Study

OBJECTIVE

People with transfemoral amputation have balance and mobility problems and are at high risk of falling. An adequate prosthetic prescription is essential to maximize their functional levels and enhance their quality of life. This study aimed to evaluate the degree of safety against falls offered by different prosthetic knees.

METHODS

A retrospective study was conducted using data from a center for prosthetic fitting and rehabilitation. Eligible individuals were adults with unilateral transfemoral amputation or knee disarticulation. The prosthetic knee models were grouped into 4 categories: locked knees, articulating mechanical knees (AMKs), fluid-controlled knees (FK), and microprocessor-controlled knees (MPK). The outcome was the number of falls experienced during inpatient rehabilitation while wearing the prosthesis. Association analyses were performed with mixed-effect Poisson models. Propensity score weighting was used to adjust causal estimates for participant confounding factors.

RESULTS

Data on 1486 hospitalizations of 815 individuals were analyzed. Most hospitalizations (77.4%) were related to individuals with amputation due to trauma. After propensity score weighting, the knee category was significantly associated with falls. People with FK had the highest rate of falling (incidence rate = 2.81 falls per 1000 patient days, 95% CI = 1.96 to 4.02). FK significantly increased the risk of falling compared with MPK (incidence rate ratio [IRRFK-MPK] = 2.44, 95% CI = 1.20 to 4.96). No other comparison among knee categories was significant.

CONCLUSIONS

Fluid-controlled prosthetic knees expose inpatients with transfemoral amputation to higher incidence of falling than MPK during rehabilitation training.

IMPACT

These findings can guide clinicians in the selection of safe prostheses and reduction of falls in people with transfemoral amputation during inpatient rehabilitation.

Propensity score analysis of the association between maternal exposure to second-hand tobacco smoke and birth defects in Northwestern China

Previous studies have suggested that maternal active smoking can increase the risk of birth defects, but evidence on second-hand tobacco smoke (SHS) is limited. We aimed to assess the association between maternal exposure to SHS and birth defects in a Chinese population. The data were based on a large-scale cross-sectional survey conducted in Shaanxi Province, China. Considering the characteristics of survey design and the potential impact of confounding factors, we adopted propensity score matching (PSM) to match the SHS exposure group and the non-exposure group to attain a balance of the confounders between the two groups. Subsequently, conditional logistic regression was employed to estimate the effect of SHS exposure on birth defects. Furthermore, sensitivity analyses were conducted to verify the key findings. After nearest neighbor matching of PSM with a ratio of 2 and a caliper width of 0.03, there were 6,205 and 12,410 participants in the exposure and control group, respectively. Pregnant women exposed to SHS were estimated to be 58% more likely to have infants with overall birth defects (OR = 1.58, 95% CI: 1.30-1.91) and 75% more likely to have infants with circulatory system defects (OR = 1.75, 95% CI: 1.26-2.44). We also observed that the risk effect of overall birth defects had an increasing trend as the frequency of exposure increased. Additionally, sensitivity analyses suggested that our results had good robustness. These results indicate that maternal exposure to SHS likely increases the risk of overall birth defects, especially circulatory system defects, in Chinese offspring.

Comparative Safety and Effectiveness of Vedolizumab to Tumor Necrosis Factor Antagonist Therapy for Ulcerative Colitis

BACKGROUND & AIMS

We aimed to compare safety and effectiveness of vedolizumab to tumor necrosis factor (TNF)-antagonist therapy in ulcerative colitis in routine practice.

METHODS

A multicenter, retrospective, observational cohort study (May 2014 to December 2017) of ulcerative colitis patients treated with vedolizumab or TNF-antagonist therapy. Propensity score weighted comparisons for development of serious adverse events and achievement of clinical remission, steroid-free clinical remission, and steroid-free deep remission. A priori determined subgroup comparisons in TNF-antagonist-naïve and -exposed patients, and for vedolizumab against infliximab and subcutaneous TNF-antagonists separately.

RESULTS

A total of 722 (454 vedolizumab, 268 TNF antagonist) patients were included. Vedolizumab-treated patients were more likely to achieve clinical remission (hazard ratio [HR], 1.651; 95% confidence interval [CI], 1.229-2.217), steroid-free clinical remission (HR, 1.828; 95% CI, 1.135-2.944), and steroid-free deep remission (HR, 2.819; 95% CI, 1.496-5.310) than those treated with TNF antagonists. Results were consistent across subgroup analyses in TNF-antagonist-naïve and -exposed patients, and for vedolizumab vs infliximab and vs subcutaneous TNF-antagonist agents separately. Overall, there were no statistically significant differences in the risk of serious adverse events (HR, 0.899; 95% CI, 0.502-1.612) or serious infections (HR, 1.235; 95% CI, 0.608-2.511) between vedolizumab-treated and TNF-antagonist-treated patients. However, in TNF-antagonist-naïve patients, vedolizumab was less likely to be associated with serious adverse events than TNF antagonists (HR, 0.192; 95% CI, 0.049-0.754).

CONCLUSIONS

Treatment of ulcerative colitis with vedolizumab is associated with higher rates of remission than treatment with TNF-antagonist therapy in routine practice, and lower rates of serious adverse events in TNF-antagonist-naïve patients.

A Comparison of Confounder Selection and Adjustment Methods for Estimating Causal Effects Using Large Healthcare Databases

PURPOSE

Confounding adjustment is required to estimate the effect of an exposure on an outcome in observational studies. However, variable selection and unmeasured confounding are particularly challenging when analyzing large healthcare data. Machine learning methods may help address these challenges. The objective was to evaluate the capacity of such methods to select confounders and reduce unmeasured confounding bias.

METHODS

A simulation study with known true effects was conducted. Completely synthetic and partially synthetic data incorporating real large healthcare data were generated. We compared Bayesian Adjustment for Confounding, Generalized Bayesian Causal Effect Estimation, Group Lasso and Doubly Robust Estimation, high-dimensional propensity score, and scalable collaborative targeted maximum likelihood algorithms. For the high-dimensional propensity score, two adjustment approaches targeting the effect in the whole population were considered: full matching and inverse probability weighting.

RESULTS

In scenarios without hidden confounders, most methods were essentially unbiased. The bias and variance of the high-dimensional propensity score varied considerably according to the number of variables selected by the algorithm. In scenarios with hidden confounders, substantial bias reduction was achieved by using machine learning methods to identify proxies as compared to adjusting only by observed confounders. High-dimensional propensity score and Group Lasso performed poorly in the partially synthetic simulation. Bayesian Adjustment for Confounding, Generalized Bayesian Causal Effect Estimation, and scalable collaborative targeted maximum likelihood algorithms performed particularly well.

CONCLUSIONS

Machine learning can help to identify measured confounders in large healthcare databases. They can also capitalize on proxies of unmeasured confounders to substantially reduce residual confounding bias. This article is protected by copyright. All rights reserved.

Suicidality risk among adults with binge-eating disorder

OBJECTIVE

To estimate relative suicidality risk associated with binge-eating disorder (BED).

METHODS

Retrospective study of patients identified as having BED (N = 1042) and a matched general population cohort (N = 10,420) from the Optum electronic health record database between January 2009 and September 2015. Patients had ≥1 outpatient encounter with a provider who recognized BED during the 12-month baseline preceding entry date. Incidence and relative risk of suicidality were assessed.

RESULTS

Incidence per 1000 person-years (95% CI) of suicidal ideation and suicide attempts, respectively, was 31.1 (23.1, 41.0) and 12.7 (7.9, 19.4) in the BED cohort and 5.8 (4.7, 7.1) and 1.4 (0.9, 2.2) in the comparator cohort. Risk of suicidal ideation and suicide attempts was greater in the BED cohort (HR [95% CIs], 6.43 [4.42, 9.37]) than in the comparator cohort (HR [95% CI], 9.47 [4.99, 17.98]) during follow-up. After adjusting for psychiatric comorbidities, associations of suicidal ideation and suicide attempts with BED remained elevated in patients with BED having histories of suicidality.

CONCLUSIONS

Findings suggest that history of suicidality may result in an increased risk of suicidal ideation and suicide attempts in patients with BED relative to the general population. Psychiatric comorbidity burden may explain the elevated risk of these conditions in BED.

Management of Gastroschisis: Results From the NETS2G Study, a Joint British, Irish, and Canadian Prospective Cohort Study of 1268 Infants

OBJECTIVE

In infants with gastroschisis, outcomes were compared between those where operative reduction and fascial closure were attempted ≤24 hours of age (PC), and those who underwent planned closure of their defect >24 hours of age following reduction with a pre-formed silo (SR).

SUMMARY OF BACKGROUND DATA

Inadequate evidence exists to determine how best to treat infants with gastroschisis.

METHODS

A secondary analysis was conducted of data collected 2006-2008 using the British Association of Pediatric Surgeons Congenital Anomalies Surveillance System, and 2005-2016 using the Canadian Pediatric Surgery Network.28-day outcomes were compared between infants undergoing PC and SR. Primary outcome was number of gastrointestinal complications. Interactions were investigated between infant characteristics and treatment to determine whether intervention effect varied in sub-groups of infants.

RESULTS

Data from 341 British and Irish infants (27%) and 927 Canadian infants (73%) were used. 671 infants (42%) underwent PC and 597 (37%) underwent SR. The effect of SR on outcome varied according to the presence/absence of intestinal perforation, intestinal matting and intestinal necrosis. In infants without these features, SR was associated with fewer gastrointestinal complications [aIRR 0.25 (95% CI 0.09-0.67, P = 0.006)], more operations [aIRR 1.40 (95% CI 1.22-1.60, P < 0.001)], more days PN [aIRR 1.08 (95% CI 1.03-1.13, P < 0.001)], and a higher infection risk [aOR 2.06 (95% CI 1.10-3.87, P = 0.025)]. In infants with these features, SR was associated with a greater number of operations [aIRR 1.30 (95% CI 1.17-1.45, P < 0.001)], and more days PN [aIRR 1.06 (95% CI 1.02-1.10, P = 0.003)].

CONCLUSIONS

In infants without intestinal perforation, matting, or necrosis, the benefits of SR outweigh its drawbacks. In infants with these features, the opposite is true. Treatment choice should be based upon these features.

Prognostic model to identify and quantify risk factors for mortality among hospitalised patients with COVID-19 in the USA

OBJECTIVES

To develop a prognostic model to identify and quantify risk factors for mortality among patients admitted to the hospital with COVID-19.

DESIGN

Retrospective cohort study. Patients were randomly assigned to either training (80%) or test (20%) sets. The training set was used to fit a multivariable logistic regression. Predictors were ranked using variable importance metrics. Models were assessed by C-indices, Brier scores and calibration plots in the test set.

SETTING

Optum de-identified COVID-19 Electronic Health Record dataset including over 700 hospitals and 7000 clinics in the USA.

PARTICIPANTS

17 086 patients hospitalised with COVID-19 between 20 February 2020 and 5 June 2020.

MAIN OUTCOME MEASURE

All-cause mortality while hospitalised.

RESULTS

The full model that included information on demographics, comorbidities, laboratory results, and vital signs had good discrimination (C-index=0.87) and was well calibrated, with some overpredictions for the most at-risk patients. Results were similar on the training and test sets, suggesting that there was little overfitting. Age was the most important risk factor. The performance of models that included all demographics and comorbidities (C-index=0.79) was only slightly better than a model that only included age (C-index=0.76). Across the study period, predicted mortality was 1.3% for patients aged 18 years old, 8.9% for 55 years old and 28.7% for 85 years old. Predicted mortality across all ages declined over the study period from 22.4% by March to 14.0% by May.

CONCLUSION

Age was the most important predictor of all-cause mortality, although vital signs and laboratory results added considerable prognostic information, with oxygen saturation, temperature, respiratory rate, lactate dehydrogenase and white cell count being among the most important predictors. Demographic and comorbidity factors did not improve model performance appreciably. The full model had good discrimination and was reasonably well calibrated, suggesting that it may be useful for assessment of prognosis.

Using propensity scores to estimate effects of treatment initiation decisions: State of the science

Confounding can cause substantial bias in nonexperimental studies that aim to estimate causal effects. Propensity score methods allow researchers to reduce bias from measured confounding by summarizing the distributions of many measured confounders in a single score based on the probability of receiving treatment. This score can then be used to mitigate imbalances in the distributions of these measured confounders between those who received the treatment of interest and those in the comparator population, resulting in less biased treatment effect estimates. This methodology was formalized by Rosenbaum and Rubin in 1983 and, since then, has been used increasingly often across a wide variety of scientific disciplines. In this review article, we provide an overview of propensity scores in the context of real-world evidence generation with a focus on their use in the setting of single treatment decisions, that is, choosing between two therapeutic options. We describe five aspects of propensity score analysis: alignment with the potential outcomes framework, implications for study design, estimation procedures, implementation options, and reporting. We add context to these concepts by highlighting how the types of comparator used, the implementation method, and balance assessment techniques have changed over time. Finally, we discuss evolving applications of propensity scores.

Comparing Propensity Score Methods Versus Traditional Regression Analysis for the Evaluation of Observational Data: A Case Study Evaluating the Treatment of Gram-Negative Bloodstream Infections

BACKGROUND

Propensity score methods are increasingly being used in the infectious diseases literature to estimate causal effects from observational data. However, there remains a general gap in understanding among clinicians on how to critically review observational studies that have incorporated these analytic techniques.

METHODS

Using a cohort of 4967 unique patients with Enterobacterales bloodstream infections, we sought to answer the question "Does transitioning patients with gram-negative bloodstream infections from intravenous to oral therapy impact 30-day mortality?" We conducted separate analyses using traditional multivariable logistic regression, propensity score matching, propensity score inverse probability of treatment weighting, and propensity score stratification using this clinical question as a case study to guide the reader through (1) the pros and cons of each approach, (2) the general steps of each approach, and (3) the interpretation of the results of each approach.

RESULTS

2161 patients met eligibility criteria with 876 (41%) transitioned to oral therapy while 1285 (59%) remained on intravenous therapy. After repeating the analysis using the 4 aforementioned methods, we found that the odds ratios were broadly similar, ranging from 0.84-0.95. However, there were some relevant differences between the interpretations of the findings of each approach.

CONCLUSIONS

Propensity score analysis is overall a more favorable approach than traditional regression analysis when estimating causal effects using observational data. However, as with all analytic methods using observational data, residual confounding will remain; only variables that are measured can be accounted for. Moreover, propensity score analysis does not compensate for poor study design or questionable data accuracy.

Doubly robust inference when combining probability and non-probability samples with high dimensional data

We consider integrating a non-probability sample with a probability sample which provides high dimensional representative covariate information of the target population. We propose a two-step approach for variable selection and finite population inference. In the first step, we use penalized estimating equations with folded concave penalties to select important variables and show selection consistency for general samples. In the second step, we focus on a doubly robust estimator of the finite population mean and re-estimate the nuisance model parameters by minimizing the asymptotic squared bias of the doubly robust estimator. This estimating strategy mitigates the possible first-step selection error and renders the doubly robust estimator root n consistent if either the sampling probability or the outcome model is correctly specified.

The effect of shifting medical coverage from National Health Insurance to Medical Aid type I and type II on health care utilization and out-of-pocket spending in South Korea

Background

This study examines the effects of a shift in medical coverage, from National Health Insurance (NHI) to Medical Aid (MA), on health care utilization (measured by the number of outpatient visits and length of stay; LOS) and out-of-pocket medical expenses.

Methods

Data were collected from the Korean Welfare Panel Study (2010–2016). A total of 888 MA Type I beneficiaries and 221 MA Type II beneficiaries who shifted from the NHI were included as the case group and 2664 and 663 consecutive NHI holders (1:3 propensity score-matched) were included as the control group, respectively. We used the ‘difference-in-differences’ (DiD) analysis approach to assess changes in health care utilization and medical spending by the group members.

Results

Differential average changes in outpatient visits in the MA Type I panel between the pre- and post-shift periods were significant, but differential changes in LOS were not found. Those who shifted from NHI to MA Type I had increased number of outpatient visits without changes in out-of-pocket spending, compared to consecutive NHI holder who had similar characteristics. However, this was not found for MA Type II beneficiaries.

Conclusion

Our research provides evidence that the shift in medical coverage from NHI to MA Type I increased the number of outpatient visits without increasing the out-of-pocket spending. Considering the problem of excess medical utilization by Korean MA Type I beneficiaries, further researches are required to have in-depth discussions on the appropriateness of the current cost-sharing level on MA beneficiaries.

Quantifying the bias due to observed individual confounders in causal treatment effect estimates

It is often of interest to use observational data to estimate the causal effect of a target exposure or treatment on an outcome. When estimating the treatment effect, it is essential to appropriately adjust for selection bias due to observed confounders using, for example, propensity score weighting. Selection bias due to confounders occurs when individuals who are treated are substantially different from those who are untreated with respect to covariates that are also associated with the outcome. A comparison of the unadjusted, naive treatment effect estimate with the propensity score adjusted treatment effect estimate provides an estimate of the selection bias due to these observed confounders. In this article, we propose methods to identify the observed covariate that explains the largest proportion of the estimated selection bias. Identification of the most influential observed covariate or covariates is important in resource-sensitive settings where the number of covariates obtained from individuals needs to be minimized due to cost and/or patient burden and in settings where this covariate can provide actionable information to healthcare agencies, providers, and stakeholders. We propose straightforward parametric and nonparametric procedures to examine the role of observed covariates and quantify the proportion of the observed selection bias explained by each covariate. We demonstrate good finite sample performance of our proposed estimates using a simulation study and use our procedures to identify the most influential covariates that explain the observed selection bias in estimating the causal effect of alcohol use on progression of Huntington's disease, a rare neurological disease.

Comparative Safety and Effectiveness of Direct-Acting Oral Anticoagulants Versus Warfarin: a National Cohort Study of Nursing Home Residents

BACKGROUND

Research comparing direct-acting oral anticoagulants (DOACs) to warfarin has excluded nursing home residents, a vulnerable and high-risk population.

OBJECTIVE

To compare the safety and effectiveness of DOACs versus warfarin.

DESIGN

New-user cohort study (2011-2016).

PATIENTS

US nursing home residents aged > 65 years with non-valvular atrial fibrillation enrolled in fee-for-service Medicare for > 6 months.

EXPOSURES

Initiators of DOACs (2881 apixaban, 1289 dabigatran, 3735 rivaroxaban) were 1:1 propensity matched to warfarin initiators.

MAIN MEASURES

Outcomes included ischemic stroke or transient ischemic attack (i.e., ischemic cerebrovascular event), bleeding (extracranial or intracranial), other vascular events, death, and a composite of all outcomes. Absolute rate differences (RD) and cause-specific hazard ratios (HR) with 95% confidence intervals (CI) were estimated. Subgroup analyses were performed by alignment of DOAC dosing with labeling.

KEY RESULTS

Median age (84 years), CHA₂DS₂-Vasc (5), and ATRIA risk scores (3) were similar across medications. Clinical outcome rates were similar for dabigatran and rivaroxaban users versus warfarin users. However, ischemic cerebrovascular event rates were higher among dabigatran and rivaroxaban users that received reduced dosages without an indication. Overall, apixaban users had higher ischemic cerebrovascular event rates (HR 1.86; 95% CI 1.00-3.45) and lower bleeding rates (HR 0.66; 95% CI 0.49-0.88), but outcome rates varied by dosing alignment. Mortality rates (per 100 person-years) were lower for apixaban (RDs - 9.30; 95% CI - 13.18 to - 5.42), dabigatran (RDs - 10.79; 95% CI - 14.98 to - 6.60), and rivaroxaban (RDs - 8.92; 95% CI - 12.01 to - 5.83) versus warfarin; composite outcome findings were similar.

CONCLUSIONS

Among US nursing home residents, the DOACs were each associated with lower mortality versus warfarin. Misaligned DOAC dosing was common in nursing homes and was associated with clinical and mortality outcomes. Overall, DOAC users had lower rates of adverse outcomes including mortality compared with warfarin users.

Comparative safety and effectiveness of vedolizumab to tumour necrosis factor antagonist therapy for Crohn's disease

BACKGROUND

Direct comparisons are lacking between vedolizumab and tumour necrosis factor (TNF)-antagonist therapy in Crohn's disease (CD).

AIM

To compare safety and effectiveness of vedolizumab and TNF-antagonist therapy in adult CD patients.

METHODS

Retrospective observational cohort (May 2014-December 2017) propensity score-weighted comparison of vedolizumab vs TNF-antagonist therapy (infliximab, adalimumab, certolizumab) in CD. Propensity scores were weighted for age, prior treatments, disease complications, extent and severity, steroid dependence, and concomitant immunosuppressive drug use. The primary outcome was comparative risk for infections or non-infectious serious adverse events (requiring antibiotics, antivirals, antifungals, hospitalisation, or treatment discontinuation, or resulting in death). Secondary comparative effectiveness outcomes were clinical remission (resolution of CD-related symptoms), steroid-free clinical remission and endoscopic remission (absence of ulcers/erosions).

RESULTS

We included 1266 patients (n = 659 vedolizumab). Rates of non-infectious serious adverse events (odds ratio [OR] 0.072, 95% confidence interval [CI] 0.012-0.242), but not serious infections (OR 1.183, 95% CI 0.786-1.795), were significantly lower with vedolizumab vs TNF-antagonist therapy. Safety comparisons for non-infectious serious adverse events remained significant after adjusting for differences in duration of exposure. No significant difference was observed between vedolizumab and TNF-antagonist therapy for clinical remission (hazard ratio [HR] 0.932, 95% CI 0.707-1.228), steroid-free clinical remission (HR 1.250, 95% CI 0.677-2.310) or endoscopic remission (HR 0.827, 95% CI 0.595-1.151). TNF-antagonist therapy was associated with higher treatment persistence compared with vedolizumab.

CONCLUSIONS

There was a lower risk of non-infectious serious adverse events, but not serious infections, with vedolizumab vs TNF-antagonist therapy, with no significant difference for achieving disease remission.

Propensity score specification for optimal estimation of average treatment effect with binary response

Propensity score methods are commonly used in statistical analyses of observational data to reduce the impact of confounding bias in estimations of average treatment effect. While the propensity score is defined as the conditional probability of a subject being in the treatment group given that subject's covariates, the most precise estimation of average treatment effect results from specifying the propensity score as a function of true confounders and predictors only. This property has been demonstrated via simulation in multiple prior research articles. However, we have seen no theoretical explanation as to why this should be so. This paper provides that theoretical proof. Furthermore, this paper presents a method for performing the necessary variable selection by means of elastic net regression, and then estimating the propensity scores so as to obtain optimal estimates of average treatment effect. The proposed method is compared against two other recently introduced methods, outcome-adaptive lasso and covariate balancing propensity score. Extensive simulation analyses are employed to determine the circumstances under which each method appears most effective. We applied the proposed methods to examine the effect of pre-cardiac surgery coagulation indicator on mortality based on a linked dataset from a retrospective review of 1390 patient medical records at Jewish Hospital (Louisville, KY) with the Society of Thoracic Surgeons database.

Clinical outcomes of theophylline use as add-on therapy in patients with chronic obstructive pulmonary disease: A propensity score matching analysis

To examine clinical outcomes of theophylline use in patients with chronic obstructive pulmonary disease (COPD) receiving inhaled corticosteroids (ICS) and long-acting beta-2 agonists (LABA). Electronic data from five hospitals located in Northern Thailand between January 2011 and December 2015 were retrospectively collected. Propensity score (PS) matching (2:1 ratio) technique was used to minimize confounding factors. The primary outcome was overall exacerbations. Secondary outcomes were exacerbation not leading to hospital admission, hospitalization for exacerbation, hospitalization for pneumonia, and all-cause hospitalizations. Cox's proportional hazards models were used to estimate adjusted hazard ratio (aHR) and 95% confidence interval (CI). After PS matching, of 711 patients with COPD (mean age: 70.1 years; 74.4% male; 60.8% severe airflow obstruction), 474 theophylline users and 237 non-theophylline users were included. Mean follow-up time was 2.26 years. Theophylline significantly increased the risk of overall exacerbation (aHR: 1.48, 95% CI: 1.11-1.96; p = 0.008) and exacerbation not leading to hospital admission (aHR: 1.47, 95% CI: 1.06-2.03; p = 0.020). Theophylline use did not significantly increase the risk of hospitalization for exacerbation (aHR: 1.11, 95% CI: 0.79-1.58; p = 0.548), hospitalization for pneumonia (aHR: 1.28, 95% CI: 0.89-1.84; p = 0.185), and all-cause hospitalizations (aHR: 1.03, 95% CI: 0.80-1.33; p = 0.795). Theophylline use as add-on therapy to ICS and LABA might be associated with an increased risk for overall exacerbation in patients with COPD. A large-scale prospective study of theophylline use investigating both safety and efficacy is warranted.

A review of the use of propensity score diagnostics in papers published in high-ranking medical journals

Background

Propensity scores are widely used to deal with confounding bias in medical research. An incorrectly specified propensity score model may lead to residual confounding bias; therefore it is essential to use diagnostics to assess propensity scores in a propensity score analysis. The current use of propensity score diagnostics in the medical literature is unknown. The objectives of this study are to (1) assess the use of propensity score diagnostics in medical studies published in high-ranking journals, and (2) assess whether the use of propensity score diagnostics differs between studies (a) in different research areas and (b) using different propensity score methods.

Methods

A PubMed search identified studies published in high-impact journals between Jan 1st 2014 and Dec 31st 2016 using propensity scores to answer an applied medical question. From each study we extracted information regarding how propensity scores were assessed and which propensity score method was used. Research area was defined using the journal categories from the Journal Citations Report.

Results

A total of 894 papers were included in the review. Of these, 187 (20.9%) failed to report whether the propensity score had been assessed. Commonly reported diagnostics were p-values from hypothesis tests (36.6%) and the standardised mean difference (34.6%). Statistical tests provided marginally stronger evidence for a difference in diagnostic use between studies in different research areas (p = 0.033) than studies using different propensity score methods (p = 0.061).

Conclusions

The use of diagnostics in the propensity score medical literature is far from optimal, with different diagnostics preferred in different areas of medicine. The propensity score literature may improve with focused efforts to change practice in areas where suboptimal practice is most common.

Automatic variable selection for exposure-driven propensity score matching with unmeasured confounders

Multivariable model building for propensity score modeling approaches is challenging. A common propensity score approach is exposure-driven propensity score matching, where the best model selection strategy is still unclear. In particular, the situation may require variable selection, while it is still unclear if variables included in the propensity score should be associated with the exposure and the outcome, with either the exposure or the outcome, with at least the exposure or with at least the outcome. Unmeasured confounders, complex correlation structures, and non-normal covariate distributions further complicate matters. We consider the performance of different modeling strategies in a simulation design with a complex but realistic structure and effects on a binary outcome. We compare the strategies in terms of bias and variance in estimated marginal exposure effects. Considering the bias in estimated marginal exposure effects, the most reliable results for estimating the propensity score are obtained by selecting variables related to the exposure. On average this results in the least bias and does not greatly increase variances. Although our results cannot be generalized, this provides a counterexample to existing recommendations in the literature based on simple simulation settings. This highlights that recommendations obtained in simple simulation settings cannot always be generalized to more complex, but realistic settings and that more complex simulation studies are needed.

Evaluation of the US Food and Drug Administration Sentinel Analysis Tools Using a Comparator with a Different Indication: Comparing the Rates of Gastrointestinal Bleeding in Warfarin and Statin Users

BACKGROUND

The US Food and Drug Administration's Sentinel System was established to monitor safety of regulated medical products. Sentinel investigators identified known associations between drugs and adverse events to test reusable analytic tools developed for Sentinel. This test case used a comparator with a different indication.

OBJECTIVE

We tested the ability of Sentinel's reusable analytic tools to identify the known association between warfarin and gastrointestinal bleeding (GIB). Statins, expected to have no effect on GIB, were the comparator. We further explored the impact of analytic features, including matching ratio and stratifying Cox regression analyses, on matched pairs.

METHODS

This evaluation included data from 14 Sentinel Data Partners. New users of warfarin and statins, aged 18 years and older, who had not received other anticoagulants or had recent GIB were matched on propensity score using 1:1 and 1:n variable ratio matching, matching statin users with warfarin users to estimate the average treatment effect in warfarin-treated patients. We compared the risk of GIB using Cox proportional hazards regression, following patients for the duration of their observed continuous treatment or until a GIB. For the 1:1 matched cohort, we conducted analyses with and without stratification on matched pair. The variable ratio matched cohort analysis was stratified on the matched set.

RESULTS

We identified 141,398 new users of warfarin and 2,275,694 new users of statins. In analyses stratified on matched pair/set, the hazard ratios (HR) for GIB in warfarin users compared with statin users were 2.78 (95% confidence interval [CI] 2.36-3.28) in the 1:1 matched cohort and 3.10 (95% CI 2.76-3.49) in the variable ratio matched cohort. The HR was lower in the analysis of the 1:1 matched cohort not stratified by matched pair (2.22, 95% CI 1.97-2.49), and highest early in treatment. Follow-up for warfarin users tended to be shorter than for statin users.

CONCLUSIONS

This study identified the expected GIB risk with warfarin compared with statins using an analytic tool developed for Sentinel. Our findings suggest that comparators with different indications may be useful in surveillance in select circumstances. Finally, in the presence of differential censoring, stratification by matched pair may reduce the potential for bias in Cox regression analyses.

Nonalcoholic fatty liver disease increases risk of incident advanced chronic kidney disease: a propensity-matched cohort study

BACKGROUND

As the prevalence of nonalcoholic fatty liver disease (NAFLD) escalates, understanding its potential impact on the development of chronic kidney disease (CKD) is needed.

OBJECTIVE

To determine the longitudinal association of NAFLD with the development of advanced CKD in the United States.

METHODS

A retrospective cohort analysis of the Truven Health MarketScan Database (2006-2015) was conducted. We used Cox proportional hazards models to compare the risk of developing CKD stages 3-5 in patients with NAFLD versus non-NAFLD, identified by ICD-9 codes, after 1:3 propensity score (PS) matching.

RESULTS

In a cohort of 262 619 newly diagnosed patients with NAFLD and 769 878 PS (1:3)-matched non-NAFLD patients, we identified 5766 and 8655 new advanced (stage 3-5) CKD cases, respectively. The crude CKD incidence rate was 8.2 and 5.5 per 1000 person-years in NAFLD and non-NAFLD groups, respectively. In multivariable Cox model, patients with NAFLD had a 41% increased risk of developing advanced CKD compared with non-NAFLD patients [adjusted hazard ratio (aHR), 1.41; 95% confidence interval (CI), 1.36-1.46]. In the sensitivity analysis adjusting for time-varying covariates after NAFLD diagnosis, NAFLD persisted as a significant CKD risk factor (aHR, 1.58; 95% CI, 1.52-1.66) and the association remained significant when stratified by age, gender and pre-existing comorbidities. The risk of CKD increased in NAFLD with compensated cirrhosis (aHR, 1.47; 95% CI, 1.36-1.59) and decompensated cirrhosis (aHR, 2.28; 95% CI, 2.12-2.46).

CONCLUSION

Nonalcoholic fatty liver disease was independently associated with an increased risk of advanced CKD development suggesting renal function screening and regular monitoring are needed in this population.

Metformin Use Is Associated With a Lower Risk of Hospitalization for Heart Failure in Patients With Type 2 Diabetes Mellitus: a Retrospective Cohort Analysis

Background A beneficial effect of metformin on heart failure requires confirmation. Methods and Results Patients with new-onset type 2 diabetes mellitus during 1999 to 2005 were enrolled from Taiwan's National Health Insurance database and followed up from January 1, 2006, until December 31, 2011. Main analyses were conducted in an unmatched cohort (172 542 metformin ever users and 43 744 never users) and a propensity score matched-pair cohort (matched cohort I, 41 714 ever users and 41 714 never users). Hazard ratios were estimated by Cox hazard regression incorporated with the inverse probability of treatment weighting using the propensity score in the unmatched cohort and by naïve method in the matched cohort I. Results showed that the respective incidence rates of heart failure hospitalization in ever users and never users were 304.25 and 864.31 per 100 000 person-years in the unmatched cohort (hazard ratio, 0.350; 95% CI, 0.329-0.373) and were 469.66 and 817.01 per 100 000 person-years in the matched cohort I (hazard ratio, 0.571; 95% CI, 0.526-0.620). A dose-response pattern was consistently observed while estimating hazard ratios for the tertiles of cumulative duration of metformin therapy. Findings were supported by another propensity score-matched cohort created after excluding 10 potential instrumental variables in the estimation of propensity score (matched cohort II). An approximately 40% lower risk was consistently observed among ever users in different models derived from the matched cohorts I and II, but models from the matched cohort II were less subject to model misspecification. Conclusions Metformin use is associated with a lower risk of heart failure hospitalization.

Inter-hospital transfer and patient outcomes: a retrospective cohort study

BACKGROUND

Inter-hospital transfer (IHT, the transfer of patients between hospitals) occurs regularly and exposes patients to risks of discontinuity of care, though outcomes of transferred patients remains largely understudied.

OBJECTIVE

To evaluate the association between IHT and healthcare utilisation and clinical outcomes.

DESIGN

Retrospective cohort.

SETTING

CMS 2013 100 % Master Beneficiary Summary and Inpatient claims files merged with 2013 American Hospital Association data.

PARTICIPANTS

Beneficiaries≥age 65 enrolled in Medicare A and B, with an acute care hospitalisation claim in 2013 and 1 of 15 top disease categories.

MAIN OUTCOME MEASURES

Cost of hospitalisation, length of stay (LOS) (of entire hospitalisation), discharge home, 3 -day and 30- day mortality, in transferred vs non-transferred patients.

RESULTS

The final cohort consisted of 53 420 transferred patients and 53 420 propensity-score matched non-transferred patients. Across all 15 disease categories, IHT was associated with significantly higher costs, longer LOS and lower odds of discharge home. Additionally, IHT was associated with lower propensity-matched odds of 3-day and/or 30- day mortality for some disease categories (acute myocardial infarction, stroke, sepsis, respiratory disease) and higher propensity-matched odds of mortality for other disease categories (oesophageal/gastrointestinal disease, renal failure, congestive heart failure, pneumonia, renal failure, chronic obstructivepulmonary disease, hip fracture/dislocation, urinary tract infection and metabolic disease).

CONCLUSIONS

In this nationally representative study of Medicare beneficiaries, IHT was associated with higher costs, longer LOS and lower odds of discharge home, but was differentially associated with odds of early death and 30 -day mortality depending on patients' disease category. These findings demonstrate heterogeneity among transferred patients depending on the diagnosis, presenting a nuanced assessment of this complex care transition.

Birth outcomes in women who have taken adalimumab in pregnancy: A prospective cohort study

BACKGROUND

Information is needed on the safety of adalimumab when used in pregnancy for the treatment of certain autoimmune diseases.

METHODS AND FINDINGS

Between 2004 and 2016, the Organization of Teratology Information Specialists Research Center at the University of California San Diego conducted a prospective controlled observational cohort study in 602 pregnant women who had or had not taken adalimumab. Women in the adalimumab-exposed cohort had received at least one dose of the drug in the first trimester for the treatment of rheumatoid arthritis or Crohn's Disease (N = 257). Women in the disease comparison cohort had not used adalimumab in pregnancy (N = 120). Women in the healthy comparison cohort had no rheumatic or inflammatory bowel diseases (N = 225). Women and their infants were followed to one year postpartum with maternal interviews, medical records abstraction, and physical examinations. Study outcomes were major structural birth defects, minor defects, spontaneous abortion, preterm delivery, pre and post-natal growth deficiency, serious or opportunistic infections and malignancies. 42/602 (7.0%) of pregnancies were lost-to-follow-up. 22/221 (10.0%) in the adalimumab-exposed cohort had a live born infant with a major birth defect compared to 8/106 (7.5%) in the diseased unexposed cohort (adjusted odds ratio 1.10, 95% confidence interval [CI] 0.45 to 2.73). Women in the adalimumab-exposed cohort were more likely to deliver preterm compared to the healthy cohort (adjusted hazard ratio [aHR] 2.59, 95% CI 1.22 to 5.50), but not compared to the diseased unexposed cohort (aHR 0.82, 95% CI 0.66 to 7.20). No significant increased risks were noted with adalimumab exposure for any other study outcomes.

CONCLUSIONS

Adalimumab exposure in pregnancy compared to diseased unexposed pregnancies was not associated with an increased risk for any of the adverse outcomes examined. Women with rheumatoid arthritis or Crohn's Disease were at increased risk of preterm delivery, irrespective of adalimumab exposure.

Project ECHO Revisited: Propensity Score Analysis And HCV Treatment Outcomes

Propensity score analysis is a statistical approach to reduce bias often present in non-randomized observational studies. In this paper we use this method to re-analyze data from a study that assessed whether patients receiving HCV treatment from providers in Project ECHO had different clinical outcomes than patients treated by specialists from an academic medical center (UNM HCV clinic) but in which treatment assignment was not randomized. We modeled the best estimated probability of treatment assignment, and then assess differences overall SVR and SVR in patients with genotype 1 infection by treatment arm using Stabilized Inverse Probability of Treatment Weights (SIPTW). Results show that after adjustment for SIPTW, HCV treatment outcomes were significantly better for the ECHO patients compared to the UNM HCV clinic patients. Higher proportions of patients treated by primary care providers achieved SVR and SVR with genotype 1 compared to those treated at UNM HCV clinic with 15.1% and 16.3% absolute differences, respectively. These results indicate that previously published results (showing no differences) were biased, and resulted in an underestimation of the treatment effect of ECHO on HCV treatment.

Propensity Score Methods in Health Technology Assessment: Principles, Extended Applications, and Recent Advances

Randomized clinical trials (RCT) are accepted as the gold-standard approaches to measure effects of intervention or treatment on outcomes. They are also the designs of choice for health technology assessment (HTA). Randomization ensures comparability, in both measured and unmeasured pretreatment characteristics, of individuals assigned to treatment and control or comparator. However, even adequately powered RCTs are not always feasible for several reasons such as cost, time, practical and ethical constraints, and limited generalizability. RCTs rely on data collected on selected, homogeneous population under highly controlled conditions; hence, they provide evidence on efficacy of interventions rather than on effectiveness. Alternatively, observational studies can provide evidence on the relative effectiveness or safety of a health technology compared to one or more alternatives when provided under the setting of routine health care practice. In observational studies, however, treatment assignment is a non-random process based on an individual’s baseline characteristics; hence, treatment groups may not be comparable in their pretreatment characteristics. As a result, direct comparison of outcomes between treatment groups might lead to biased estimate of the treatment effect. Propensity score approaches have been used to achieve balance or comparability of treatment groups in terms of their measured pretreatment covariates thereby controlling for confounding bias in estimating treatment effects. Despite the popularity of propensity scores methods and recent important methodological advances, misunderstandings on their applications and limitations are all too common. In this article, we present a review of the propensity scores methods, extended applications, recent advances, and their strengths and limitations.

Propensity score methods to control for confounding in observational cohort studies: a statistical primer and application to endoscopy research

BACKGROUND AND AIMS

Confounding is a major concern in nonexperimental studies of endoscopic interventions and can lead to biased estimates of the effects of treatment. Propensity score methods, which are commonly used in the pharmacoepidemiology literature, can effectively control for baseline confounding by balancing measured baseline confounders and risk factors and creating comparable populations of treated and untreated patients.

METHODS

We propose the following 5-step checklist to guide the use and evaluation of propensity score methods: (1) select covariates, (2) assess "Table 1" balance in risk factors before propensity score implementation, (3) estimate and implement the propensity score in the study cohort, (4) reassess "Table 1" balance in risk factors after propensity score implementation, and (5) critically evaluate differences between matched and unmatched patients after propensity score implementation. We then applied this checklist to an endoscopy example using a study cohort of 411 adults with newly diagnosed eosinophilic esophagitis (EoE), some of whom were treated with esophageal dilation.

RESULTS

We identified 156 patients, aged 18 and older, who were treated with esophageal dilation, and 255 patients who were nondilated. We successfully matched 148 (95%) dilated patients to nondilated patients who had a propensity score within 0.1, based on patient age, sex, race, self-reported food allergy, and presence of narrowing at baseline endoscopy. Crude imbalances were observed before propensity score matching in several baseline covariates, including age, sex, and narrowing; however, propensity score matching was successful in achieving balance across all measured covariates.

CONCLUSIONS

We provide an introduction to propensity score methods, including a straightforward checklist for implementing propensity score methods in nonexperimental studies of treatment effectiveness. Moreover, we demonstrate the advantage of using "Table 1" as a simple but effective diagnostic tool for evaluating the success of propensity score methods in an applied example of esophageal dilation in EoE.

Poor compliance with perioperative chemotherapy for resectable gastric cancer and its impact on survival

BACKGROUND

In several Western European countries it is recommended to treat gastric cancer patients with perioperative chemotherapy if they are eligible for surgery. However, little is known about its use in daily clinical practice. This study examines the use of perioperative treatment and its impact on survival in the Netherlands.

METHODS

Patients diagnosed with potentially resectable gastric cancer (cT1N+/cT2-T3,X any cN, cM0,X) between 2006 and 2014 were selected from the Netherlands Cancer Registry (N = 5824). Treatment trends were examined. Propensity score matching was used to create a subsample to reduce selection bias. Cox regression analysis was used to assess differences in overall survival.

RESULTS

The percentage of patients treated with perioperative treatment increased from 3% in 2006 to 26% in 2014 and the use of only surgery decreased from 60% to 26%. 35% of all patients did not undergo surgery. Of the patients who underwent preoperative chemotherapy and surgery, 43% did not commence postoperative treatment. Cox regression analysis showed a better overall survival for patients who underwent perioperative treatment compared to patients who underwent preoperative treatment only (HR = 0.80 95%CI 0.70-0.93; propensity matched sample: HR = 0.84 95%CI 0.71-0.99), whereas survival was comparable for patients who underwent preoperative chemotherapy versus surgery alone (HR = 0.89 95%CI 0.77-1.02, propensity matched sample: HR = 0.85 95%CI 0.72-1.01).

CONCLUSION

This population-based study highlights that a significant proportion of the patients did not receive perioperative treatment. More research is necessary to elucidate the importance of the individual components of perioperative treatment.

Impact of metformin use on the cardiovascular effects of dipeptidyl peptidase-4 inhibitors: An analysis of Medicare claims data from 2007 to 2015

AIMS

To examine the outcomes of dipeptidyl peptidase-4 (DPP-4) inhibitor initiation with and without concurrent metformin treatment.

MATERIALS AND METHODS

We identified Medicare enrollees initiating a DPP-4 inhibitor, a sulphonylurea or a thiazolidinedione. Using propensity-score-weighted Poisson models, we evaluated 1-year cardiovascular (CV) outcome incidence among initiators of DPP-4 inhibitors versus comparators in subgroups with and without concurrent metformin use, and assessed the interaction between initiation drug and metformin. Outcomes included mortality, non-fatal myocardial infarction (MI), stroke, and a composite outcome.

RESULTS

For the DPP-4 inhibitor (n = 13 391) versus sulphonylurea (n = 33 206) comparison, rate differences in composite outcome incidence favoured DPP-4 inhibitors: -2.0/100 person-years among metformin users (95% confidence interval [CI] -2.7 to -1.3) and - 1.0/100 person-years (95% CI -1.8 to -0.2) among metformin non-users. Similar rate difference trends among metformin users and non-users were seen for mortality (-1.5/100 person-years [95% CI -2.1 to -0.9] and -0.7/100 person-years [95% CI -1.4 to 0.0]) and non-fatal MI (-0.5/100 person-years [95% CI -0.8, -0.3] and 0.1/100 person-years [95% CI -0.2 to 0.4]). The interaction between DPP-4 inhibitor initiation and metformin was statistically significant for non-fatal MI (P = 0.008). For the DPP-4 inhibitor (n = 22 210) versus thiazolidinedione (n = 9517) comparison, rate differences in composite outcome incidence for DPP-4 inhibitor initiation were -0.6/100 person-years (95% CI -1.5 to 0.2) among metformin users and 1.0 (95% CI 0.0 to 2.0) among metformin non-users. Similar rate difference trends among metformin users and non-users were seen for mortality (-0.5/100 person-years [95% CI -1.3 to 0.1] and 0.8/100 person-years [95% CI -0.0 to 1.7]) and non-fatal MI (-0.1/100 person-years [95% CI -0.4 to 0.2] and 0.2/100 person-years [95% CI -0.1 to 0.6]). The interaction between DPP-4 inhibitor initiation and metformin was statistically significant for the composite outcome (P = 0.024) and mortality (P = 0.023).

CONCLUSION

Incidence rate differences in multiple CV outcomes appeared more favourable when DPP-4 inhibitor initiation occurred in the presence of metformin, suggesting a possible interaction between DPP-4 inhibitors and metformin.

The importance and implications of comparator selection in pharmacoepidemiologic research

PURPOSE OF REVIEW

Pharmacoepidemiologic studies employing large databases are critical to evaluating the effectiveness and safety of drug exposures in large and diverse populations. Because treatment is not randomized, researchers must select a relevant comparison group for the treatment of interest. The comparator group can consist of individuals initiating: (1) a similarly indicated treatment (active comparator), (2) a treatment used for a different indication (inactive comparator) or (3) no particular treatment (non-initiators). Herein we review recent literature and describe considerations and implications of comparator selection in pharmacoepidemiologic studies.

RECENT FINDINGS

Comparator selection depends on the scientific question and feasibility constraints. Because pharmacoepidemiologic studies rely on the choice to initiate or not initiate a specific treatment, rather than randomization, they are at-risk for confounding related to the comparator choice including: by indication, disease severity and frailty. We describe forms of confounding specific to pharmacoepidemiologic studies and discuss each comparator along with informative examples and a case study. We provide commentary on potential issues relevant to comparator selection in each study, highlighting the importance of understanding the population in whom the treatment is given and how patient characteristics are associated with the outcome.

SUMMARY

Advanced statistical techniques may be insufficient for reducing confounding in observational studies. Evaluating the extent to which comparator selection may mitigate or induce systematic bias is a critical component of pharmacoepidemiologic studies.

Propensity Score-Based Methods in Comparative Effectiveness Research on Coronary Artery Disease

This review examines the conduct and reporting of observational studies using propensity score-based methods to compare coronary artery bypass grafting (CABG), percutaneous coronary intervention (PCI), or medical therapy for patients with coronary artery disease. A systematic selection process identified 48 studies: 20 addressing CABG versus PCI; 21 addressing bare-metal stents versus drug-eluting stents; 5 addressing CABG versus medical therapy; 1 addressing PCI versus medical therapy; and 1 addressing drug-eluting stents versus balloon angioplasty. Of 32 studies reporting information on variable selection, 7 relied exclusively on statistical criteria for the association of covariates with treatment, and 5 used such criteria to determine whether product or nonlinear terms should be included in the propensity score model. Twenty-five (52%) studies reported assessing covariate balance using the estimated propensity score, but only 1 described modifications to the propensity score model based on this assessment. The over 400 variables used in the 48 propensity score models were classified into 12 categories and 60 subcategories; only 17 subcategories were represented in at least half of the propensity score models. Overall, reporting of propensity score-based methods in observational studies comparing CABG, PCI, and medical therapy was incomplete; when adequately described, the methods used were often inconsistent with current methodological standards.

Analysis of causality from observational studies and its application in clinical research in Intensive Care Medicine

Random allocation of treatment or intervention is the key feature of clinical trials and divides patients into treatment groups that are approximately balanced for baseline, and therefore comparable covariates except for the variable treatment of the study. However, in observational studies, where treatment allocation is not random, patients in the treatment and control groups often differ in covariates that are related to intervention variables. These imbalances in covariates can lead to biased estimates of the treatment effect. However, randomized clinical trials are sometimes not feasible for ethical, logistical, economic or other reasons. To resolve these situations, interest in the field of clinical research has grown in designing studies that are most similar to randomized experiments using observational (i.e. non-random) data. Observational studies using propensity score analysis methods have been increasing in the scientific papers of Intensive Care. Propensity score analyses attempt to control for confounding in non-experimental studies by adjusting for the likelihood that a given patient is exposed. However, studies with propensity indexes may be confusing, and intensivists are not familiar with this methodology and may not fully understand the importance of this technique. The objectives of this review are: to describe the fundamentals of propensity index methods; to present the techniques to adequately evaluate propensity index models; to discuss the advantages and disadvantages of these techniques.

Chronic hepatitis C virus (HCV) increases the risk of chronic kidney disease (CKD) while effective HCV treatment decreases the incidence of CKD

We assessed the risk of chronic kidney disease (CKD) in chronic hepatitis C virus (HCV)-infected patients and the incidence reduction of CKD after receipt of HCV treatment. We also evaluated the risk of membranoproliferative glomerulonephritis (MPGN) and cryoglobulinemia in chronic HCV patients. A retrospective cohort analysis of the Truven Health MarketScan Database (2008-2015) in the United States was conducted. In a cohort of 56,448 HCV-infected patients and 169,344 propensity score (1:3)-matched non-HCV patients, we examined the association of HCV infection with the incidence of CKD. Of 55,818 HCV patients, 6.6 % (n = 3666), 6.3% (n = 3534), and 8.3% (n = 4628) patients received either interferon-based dual, triple, or all-oral direct acting antiviral agent therapy, respectively, whereas 79% of patients did not receive any HCV treatment. Cox proportional hazards models were used to compare the risk of developing CKD in HCV patients compared with non-HCV patients and treated patients compared with untreated HCV patients. In a multivariate time-varying Cox regression model, HCV-infected patients had a 27% increased risk of CKD compared with non-HCV patients (hazard ratio [HR], 1.27; 95% confidence interval [CI], 1.18-1.37). Among HCV patients, individuals who received the minimally effective HCV treatment for dual, triple, or all-oral therapy had a 30% decreased risk of developing CKD (HR, 0.70; 95% CI, 0.55-0.88). In addition, HCV-infected patients experienced a twofold and a nearly 17-fold higher risk of MPGN (HR, 2.23; 95% CI, 1.84-2.71) and cryoglobulinemia (HR, 16.91; 95% CI, 12.00-23.81) respectively, compared with non-HCV patients. Conclusion: HCV-infected individuals in the United States are at greater risk of developing CKD, MPGN, and cryoglobulinemia. Minimally effective treatment of HCV infection can prevent the development of CKD, although the association was not significant for all-oral therapy. (Hepatology 2018;67:492-504).

The effect of adult children living in the United States on the likelihood of cognitive impairment for older parents living in Mexico

OBJECTIVE

The increased risk for poor physical and mental health outcomes for older parents in Mexico who have an adult child living in the United States may contribute to an increased risk for cognitive impairment in this population. The objective of this study was to examine if older adults in Mexico who have one or more adult children living in the United States are more or less likely to develop cognitive impairment over an 11-year period compared to older adults who do not have any adult children living in the United States.

DESIGN

Data for this study came from Wave I (2001) and Wave III (2012) of the Mexican Health and Aging Study. The final sample included 2609 participants aged 60 and over who were not cognitively impaired in 2001 and had one or more adult children (age ≥15). Participants were matched using a propensity score that was estimated with a multivariable logistic regression model that included sociodemographic characteristics and migration history of the older parents.

RESULTS

Having one or more adult children living in the United States is associated with lower socioeconomic status and higher number of depressive symptoms, but greater social engagement for older parents living in Mexico. No significant differences in the odds for developing cognitive impairment according to having one or more adult children living in the United States were detected.

CONCLUSION

In summary, having one or more adult children living in the United States was associated with characteristics that may increase and decrease the risk for cognitive impairment. This may contribute to the non-significant relationship between migration status of adult children and likelihood for cognitive impairment for older parents living in Mexico.

Outcome-adaptive lasso: Variable selection for causal inference

Methodological advancements, including propensity score methods, have resulted in improved unbiased estimation of treatment effects from observational data. Traditionally, a "throw in the kitchen sink" approach has been used to select covariates for inclusion into the propensity score, but recent work shows including unnecessary covariates can impact both the bias and statistical efficiency of propensity score estimators. In particular, the inclusion of covariates that impact exposure but not the outcome, can inflate standard errors without improving bias, while the inclusion of covariates associated with the outcome but unrelated to exposure can improve precision. We propose the outcome-adaptive lasso for selecting appropriate covariates for inclusion in propensity score models to account for confounding bias and maintaining statistical efficiency. This proposed approach can perform variable selection in the presence of a large number of spurious covariates, that is, covariates unrelated to outcome or exposure. We present theoretical and simulation results indicating that the outcome-adaptive lasso selects the propensity score model that includes all true confounders and predictors of outcome, while excluding other covariates. We illustrate covariate selection using the outcome-adaptive lasso, including comparison to alternative approaches, using simulated data and in a survey of patients using opioid therapy to manage chronic pain.

Antipsychotic Use Among Adult Outpatients and Venous Thromboembolic Disease: A Retrospective Cohort Study

BACKGROUND

Treatment with antipsychotic (AP) agents is associated with incident thromboembolic events. However, the underpinnings of this association remain unknown. We sought to evaluate the effect of AP agents-categorized by their metabolic/sedative and hyperprolactinemia adverse effect profile-on the risk of venous thromboembolic disease during long-term follow-up.

METHODS

A retrospective cohort study of adult patients initiating AP treatment for the first time was conducted. Primary outcome was defined as the time to venous thromboembolism (VTE) (either deep venous thrombosis or acute pulmonary embolism). Antipsychotic agents were categorized by their risk (high vs low) of either drug-induced (a) sedation/metabolic adverse event or (b) hyperprolactinemia. We used a propensity score-adjusted Cox proportional hazards model to control for confounding.

FINDINGS

One thousand eight patients (mean age, 72.4 y) were followed for a median of 36 months. Incident VTE occurred in 6.25% of patients, corresponding to an incidence rate of 184 cases per 10,000 person-years. We found no difference in the hazard of VTE during follow-up between high- and low-risk groups (hazard ratio, 1.23 [95% confidence interval, 0.74-2.04] for drug-induced sedation/metabolic adverse event risk categories, and hazard ratio 0.81 [95% confidence interval, 0.50-1.35] for high versus low hyperprolactinemia risk).

CONCLUSIONS

These results suggest that the risk of thromboembolic events in older adults who started AP treatment for the first time does not seem to be related to these drugs' risk of either sedation/metabolic adverse events or hyperprolactinemia. However, VTE remains a common problem in this subgroup of patients.

The feasibility of matching on a propensity score for acupuncture in a prospective cohort study of patients with chronic pain

Background

Propensity scores are typically applied in retrospective cohort studies. We describe the feasibility of matching on a propensity score derived from a retrospective cohort and subsequently applied in a prospective cohort study of patients with chronic musculoskeletal pain before the start of acupuncture or usual care treatment and enrollment in a comparative effectiveness study that required patient reported pain outcomes.

Methods

We assembled a retrospective cohort study using data from 2010 to develop a propensity score for acupuncture versus usual care based on electronic healthcare record and administrative data (e.g., pharmacy) from an integrated health plan, Kaiser Permanente Northwest. The propensity score’s probabilities allowed us to match acupuncture-referred and non-referred patients prospectively in 2013-14 after a routine outpatient visit for pain. Among the matched patients, we collected patient-reported pain before treatment and during follow-up to assess the comparative effectiveness of acupuncture. We assessed balance in patient characteristics with the post-matching c-statistic and standardized differences.

Results

Based on the propensity score and other characteristics (e.g., patient-reported pain), we were able to match all 173 acupuncture-referred patients to 350 non-referred (usual care) patients. We observed a residual imbalance (based on the standardized differences) for some characteristics that contributed to the score; for example, age, -0.283, and the Charlson comorbidity score, -0.264, had the largest standardized differences. The overall balance of the propensity score appeared more favorable according to the post-matching c-statistic, 0.503.

Conclusion

The propensity score matching was feasible statistically and logistically and allowed approximate balance on patient characteristics, some of which will require adjustment in the comparative effectiveness regression model. By transporting propensity scores to new patients, healthcare systems with electronic health records can conduct comparative effectiveness cohort studies that require prospective data collection, such as patient-reported outcomes, while approximately balancing numerous patient characteristics that might confound the benefit of an intervention. The approach offers a new study design option.