Longitudinal histories as predictors of future diagnoses of domestic abuse: modelling study

Machine learning-based algorithms applied to drug prescriptions and other healthcare services in the Sicilian claims database to identify acromegaly as a model for the earlier diagnosis of rare diseases

Acromegaly is a rare disease characterized by a diagnostic delay ranging from 5 to 10 years from the symptoms' onset. The aim of this study was to develop and internally validate machine-learning algorithms to identify a combination of variables for the early diagnosis of acromegaly. This retrospective population-based study was conducted between 2011 and 2018 using data from the claims databases of Sicily Region, in Southern Italy. To identify combinations of potential predictors of acromegaly diagnosis, conditional and unconditional penalized multivariable logistic regression models and three machine learning algorithms (i.e., the Recursive Partitioning and Regression Tree, the Random Forest and the Support Vector Machine) were used, and their performance was evaluated. The random forest (RF) algorithm achieved the highest Area under the ROC Curve value of 0.83 (95% CI 0.79-0.87). The sensitivity in the test set, computed at the optimal threshold of predicted probabilities, ranged from 28% for the unconditional logistic regression model to 69% for the RF. Overall, the only diagnosis predictor selected by all five models and algorithms was the number of immunosuppressants-related pharmacy claims. The other predictors selected by at least two models were eventually combined in an unconditional logistic regression to develop a meta-score that achieved an acceptable discrimination accuracy (AUC = 0.71, 95% CI 0.66-0.75). Findings of this study showed that data-driven machine learning algorithms may play a role in supporting the early diagnosis of rare diseases such as acromegaly.

Development and external validation of a pretrained deep learning model for the prediction of non-accidental trauma

Non-accidental trauma (NAT) is deadly and difficult to predict. Transformer models pretrained on large datasets have recently produced state of the art performance on diverse prediction tasks, but the optimal pretraining strategies for diagnostic predictions are not known. Here we report the development and external validation of Pretrained and Adapted BERT for Longitudinal Outcomes (PABLO), a transformer-based deep learning model with multitask clinical pretraining, to identify patients who will receive a diagnosis of NAT in the next year. We develop a clinical interface to visualize patient trajectories, model predictions, and individual risk factors. In two comprehensive statewide databases, approximately 1% of patients experience NAT within one year of prediction. PABLO predicts NAT events with area under the receiver operating characteristic curve (AUROC) of 0.844 (95% CI 0.838-0.851) in the California test set, and 0.849 (95% CI 0.846-0.851) on external validation in Florida, outperforming comparator models. Multitask pretraining significantly improves model performance. Attribution analysis shows substance use, psychiatric, and injury diagnoses, in the context of age and racial demographics, as influential predictors of NAT. As a clinical decision support system, PABLO can identify high-risk patients and patient-specific risk factors, which can be used to target secondary screening and preventive interventions at the point-of-care.

Imaging of Intimate Partner Violence, From the AJR Special Series on Emergency Radiology

Intimate partner violence (IPV) is a highly prevalent public health issue with multiple adverse health effects. Radiologists are well suited to assessing a patient's likelihood of IPV. Recognition of common IPV injury mechanisms and resulting target and defensive injury patterns on imaging and understanding of differences between patients who have experienced IPV and those who have not with respect to use of imaging will aid radiologists in accurate IPV diagnosis. Target injuries often involve the face and neck as a result of blunt trauma or strangulation; defensive injuries often involve an extremity. Awareness of differences in injury patterns resulting from IPV-related and accidental trauma can aid radiologists in detecting a mismatch between the provided clinical history and imaging findings to support suspicion of IPV. Radiologists should consider all available current and prior imaging in assessing the likelihood of IPV; this process may be aided by machine learning methods. Even if correctly suspecting IPV on the basis of imaging, radiologists face challenges in acting on that suspicion, including appropriately documenting the findings, without compromising the patient's confidentiality and safety. However, through a multidisciplinary approach with appropriate support mechanisms, radiologists may serve as effective frontline physicians for raising suspicion of IPV.

Economic Hardship and Violence: A Comparison of County-Level Economic Measures in the Prediction of Violence-Related Injury

Economic hardship may lead to a wide range of negative outcomes, including violence. However, existing literature on economic hardship and violence is limited by reliance on official reports of violence and conflation of different measures of economic hardship. The goals of this study are to measure how violence-related injuries are associated with five measures of county-level economic shocks: unemployment rate, male mass layoffs, female mass layoffs, foreclosure rate, and unemployment rate change, measured cross-sectionally and by a 1-year lag. This study measures three subtypes of violence outcomes (child abuse, elder abuse, and intimate partner violence). Yearly county-level data were obtained on violence-related injuries and economic measures from 2005 to 2012 for all 87 counties in Minnesota. Negative binomial models were run regressing the case counts of each violence outcome at the county-year level on each economic indicator modeled individually, with population denominator offsets to yield incidence rate ratios. Crude models were run first, then county-level socio-demographic variables and year were added to each model, and finally fully-adjusted models were run including all socio-demographic variables plus all economic indicators simultaneously. In the fully-adjusted models, a county's higher foreclosure rate is the strongest and most consistently associated with an increase in all violence subtypes. Unemployment rate is the second strongest and most consistent economic risk factor for all violence subtypes. Lastly, there appears to be an impact of gender specific to economic impacts on child abuse; specifically, male mass-lay-offs were associated with increased rates while female mass-lay-offs were associated with decreased rates. Understanding the associations of different types of economic hardship with a range of violence outcomes can aid in developing more holistic prevention and intervention efforts.

Imaging Findings on Head Computed Tomography Scans in Victims of Intimate Partner Violence

OBJECTIVE

The aim of the study is to analyze the imaging findings and injury patterns seen on head computed tomography (CT) examinations performed on survivors of intimate partner violence (IPV).

METHODS

An institutional review board-approved retrospective analysis of 668 patients reporting IPV to our institution's violence intervention and prevention program between January 2013 and June 2018 identified 40 unique patients with radiological findings visible on head CT. All injuries visible on head CT were analyzed based on the anatomic location and injury type. Demographics, IPV screening at the time of injury, concomitant, prior, and subsequent injuries to the index head injury were also recorded.

RESULTS

Our study cohort had 36 women and 4 men with a mean age at presentation of 43 ± 13 years (mean ± SD), 91 unique injuries with 57 (62.6%) isolated soft tissue injuries, 4 (3.2%) fractures, 13 (14.3%) intra-axial, and 17 (18.7%) extra-axial injuries. Soft tissue injuries and intra-axial injuries occurred most commonly in the frontal region (45.6% and 38.5%), followed by the parietal region (22.8% and 23.1%), while most extra-axial injuries were subdural hematomas (41.2%). Left-sided injuries accounted for 49% (45/91) with 29/91 right-sided (32%) and 17/91 bilateral (19%) injuries. The IPV screening occurred in 44% of injury visits (22/50). Concomitant injuries were seen in 14/50 injury visits (28%), most commonly being in the lower extremity (6/14, 42.9% [% of visits with concomitant injuries]) followed by the upper extremity (5/14, 35.7%), while 52% of visits (26/50) were preceded by prior injuries and 68% of events (34/50) were followed by subsequent injuries.

CONCLUSIONS

Isolated soft tissue swelling is the most common manifestation of IPV on head CT scans with frontoparietal region being the most common site. Synchronous and metachronous injuries are frequent.

Improving risk prediction for target subpopulations: Predicting suicidal behaviors among multiple sclerosis patients

Several recent studies have applied machine learning techniques to develop risk algorithms that predict subsequent suicidal behavior based on electronic health record data. In this study we used a retrospective cohort study design to test whether developing more tailored predictive models-within specific subpopulations of patients-would improve predictive accuracy. A retrospective cohort of 15,117 patients diagnosed with multiple sclerosis (MS), a diagnosis associated with increased risk of suicidal behavior, was used. The cohort was randomly divided into equal sized training and validation sets. Overall, suicidal behavior was identified among 191 (1.3%) of the patients with MS. A Naïve Bayes Classifier model was trained on the training set to predict future suicidal behavior. With 90% specificity, the model detected 37% of subjects who later demonstrated suicidal behavior, on average 4.6 years before the first suicide attempt. The performance of a model trained only on MS patients was better at predicting suicide in MS patients than that a model trained on a general patient sample of a similar size (AUC of 0.77 vs. 0.66). Unique risk factors for suicidal behavior among patients with MS included pain-related codes, gastroenteritis and colitis, and history of smoking. Future studies are needed to further test the value of developing population-specific risk models.

Violence in the Great Recession

Substantial evidence suggests that economic hardship causes violence. However, a large majority of this research relies on observational studies that use traditional violence surveillance systems that suffer from selection bias and over-represent vulnerable populations, such as people of color. To overcome limitations of prior work, we employed a quasi-experimental design to assess the impact of the Great Recession on explicit violence diagnoses (injuries identified to be caused by a violent event) and proxy violence diagnoses (injuries highly correlated with violence) for child maltreatment, intimate partner violence, elder abuse, and their combination. We used Minnesota hospital data (2004-2014), conducting a difference-in-differences analysis at the county level (n = 86) using linear regression to compare changes in violence rates from before the recession (2004-2007) to after the recession (2008-2014) in counties most affected by the recession, versus changes over the same time period in counties less affected by the recession. The findings suggested that the Great Recession had little or no impact on explicitly identified violence; however, it affected proxy-identified violence. Counties that were more highly affected by the Great Recession saw a greater increase in the average rate of proxy-identified child abuse, elder abuse, intimate partner violence, and combined violence when compared with less-affected counties.

Predictive structured-unstructured interactions in EHR models: A case study of suicide prediction

Clinical risk prediction models powered by electronic health records (EHRs) are becoming increasingly widespread in clinical practice. With suicide-related mortality rates rising in recent years, it is becoming increasingly urgent to understand, predict, and prevent suicidal behavior. Here, we compare the predictive value of structured and unstructured EHR data for predicting suicide risk. We find that Naive Bayes Classifier (NBC) and Random Forest (RF) models trained on structured EHR data perform better than those based on unstructured EHR data. An NBC model trained on both structured and unstructured data yields similar performance (AUC = 0.743) to an NBC model trained on structured data alone (0.742, p = 0.668), while an RF model trained on both data types yields significantly better results (AUC = 0.903) than an RF model trained on structured data alone (0.887, p < 0.001), likely due to the RF model’s ability to capture interactions between the two data types. To investigate these interactions, we propose and implement a general framework for identifying specific structured-unstructured feature pairs whose interactions differ between case and non-case cohorts, and thus have the potential to improve predictive performance and increase understanding of clinical risk. We find that such feature pairs tend to capture heterogeneous pairs of general concepts, rather than homogeneous pairs of specific concepts. These findings and this framework can be used to improve current and future EHR-based clinical modeling efforts.

Temporally informed random forests for suicide risk prediction

OBJECTIVE

Suicide is one of the leading causes of death worldwide, yet clinicians find it difficult to reliably identify individuals at high risk for suicide. Algorithmic approaches for suicide risk detection have been developed in recent years, mostly based on data from electronic health records (EHRs). Significant room for improvement remains in the way these models take advantage of temporal information to improve predictions.

MATERIALS AND METHODS

We propose a temporally enhanced variant of the random forest (RF) model-Omni-Temporal Balanced Random Forests (OT-BRFs)-that incorporates temporal information in every tree within the forest. We develop and validate this model using longitudinal EHRs and clinician notes from the Mass General Brigham Health System recorded between 1998 and 2018, and compare its performance to a baseline Naive Bayes Classifier and 2 standard versions of balanced RFs.

RESULTS

Temporal variables were found to be associated with suicide risk: Elevated suicide risk was observed in individuals with a higher total number of visits as well as those with a low rate of visits over time, while lower suicide risk was observed in individuals with a longer period of EHR coverage. RF models were more accurate than Naive Bayesian classifiers at predicting suicide risk in advance (area under the receiver operating curve = 0.824 vs. 0.754, respectively). The proposed OT-BRF model performed best among all RF approaches, yielding a sensitivity of 0.339 at 95% specificity, compared to 0.290 and 0.286 for the other 2 RF models. Temporal variables were assigned high importance by the models that incorporated them.

DISCUSSION

We demonstrate that temporal variables have an important role to play in suicide risk detection and that requiring their inclusion in all RF trees leads to increased predictive performance. Integrating temporal information into risk prediction models helps the models interpret patient data in temporal context, improving predictive performance.

Longitudinal imaging history in early identification of intimate partner violence

OBJECTIVES

To describe the imaging findings of intimate partner violence (IPV)-related injury and to evaluate the role of longitudinal imaging review in detecting IPV.

METHODS

Radiology studies were reviewed in chronological order and IPV-related injuries were recorded among 400 victims of any type of abuse (group 1) and 288 of physical abuse (group 2) from January 2013 to June 2018. The likelihood of IPV was assessed as low/moderate/high based on the review of (1) current and prior anatomically related studies only and (2) longitudinal imaging history consisting of all prior studies. The first radiological study date with moderate/high suspicion was compared to the self-reported date by the victim.

RESULTS

A total of 135 victims (33.8%) in group 1 and 144 victims (50%) in group 2 demonstrated IPV-related injuries. Musculoskeletal injury was most common (58.2% and 44.5% in groups 1 and 2, respectively; most commonly lower/upper extremity fractures), followed by neurologic injury (20.9% and 32.9% in groups 1 and 2, respectively; most commonly facial injury). With longitudinal imaging history, radiologists were able to identify IPV in 31% of group 1 and 46.5% of group 2 patients. Amongst these patients, earlier identification by radiologists was provided compared to the self-reported date in 62.3% of group 1 (median, 64 months) and in 52.6% of group 2 (median, 69.3 months).

CONCLUSIONS

Musculoskeletal and neurological injuries were the most common IPV-related injuries. Knowledge of common injuries and longitudinal imaging history may help IPV identification when victims are not forthcoming.

KEY POINTS

• Musculoskeletal injuries were the most common type of IPV-related injury, followed by neurological injuries. • With longitudinal imaging history, radiologists were able to better raise the suspicion of IPV compared to the selective review of anatomically related studies only. • With longitudinal imaging history, radiologists were able to identify IPV earlier than the self-reported date by a median of 64 months in any type of abuse, and a median of 69.3 months in physical abuse.

Measuring the hidden burden of violence: use of explicit and proxy codes in Minnesota injury hospitalizations, 2004-2014

PURPOSE

Commonly-used violence surveillance systems are biased towards certain populations due to overreporting or over-scrutinized. Hospital discharge data may offer a more representative view of violence, through use of proxy codes, i.e. diagnosis of injuries correlated with violence. The goals of this paper are to compare the trends in violence in Minnesota, and associations of county-level demographic characteristics with violence rates, measured through explicitly diagnosed violence and proxy codes. It is an exploration of how certain sub-populations are overrepresented in traditional surveillance systems.

METHODS

Using Minnesota hospital discharge data linked with census data from 2004 to 2014, this study examined the distribution and time trends of explicit, proxy, and combined (proxy and explicit) codes for child abuse, intimate partner violence (IPV), and elder abuse. The associations between county-level risk factors (e.g., poverty) and county violence rates were estimated using negative binomial regression models with generalized estimation equations to account for clustering over time.

RESULTS

The main finding was that the patterns of county-level violence differed depending on whether one used explicit or proxy codes. In particular, explicit codes suggested that child abuse and IPV trends were flat or decreased slightly from 2004 to 2014, while proxy codes suggested the opposite. Elder abuse increased during this timeframe for both explicit and proxy codes, but more dramatically when using proxy codes. In regard to the associations between county level characteristics and each violence subtype, previously identified county-level risk factors were more strongly related to explicitly-identified violence than to proxy-identified violence. Given the larger number of proxy-identified cases as compared with explicit-identified violence cases, the trends and associations of combined codes align more closely with proxy codes, especially for elder abuse and IPV.

CONCLUSIONS

Violence surveillance utilizing hospital discharge data, and particularly proxy codes, may add important information that traditional surveillance misses. Most importantly, explicit and proxy codes indicate different associations with county sociodemographic characteristics. Future research should examine hospital discharge data for violence identification to validate proxy codes that can be utilized to help to identify the hidden burden of violence.

Upper extremity fractures due to intimate partner violence versus accidental causes

Purpose

The purpose of this study is to evaluate the prevalence of intimate partner violence (IPV)-related upper extremity fractures (UEF) in women presenting to US emergency departments (ED) and compare their anatomic location to those due to accidental falls or strikes.

Methods

An Institutional Review Board exempt, retrospective review of prospectively collected data was performed using the National Electronic Injury Surveillance System’s All Injury Program data from 2005 through 2015 for all UEF sustained in women 15 to 54 years old. Injuries based on reported IPV versus accidental falls or strikes were analyzed accounting for the weighted, stratified nature of the data.

Results

IPV-related UEF represented 1.7% of all UEF and 27.2% of all IPV fractures. The finger was the most common fracture site in IPV (34.3%) and accidental striking (53.3%) but accounted for only 10% of fall-related UEF. There was a higher proportion of shoulder fractures in IPV (9.2%) compared to accidental falls (7.4%) or strikes (2.9%). The odds of a finger fracture were 4.32 times greater in IPV than falling and of a shoulder fracture were 3.65 greater in IPV than accidental striking (p < 0.0001).

Conclusions

While the finger is the most common site for IPV UEF, it is also the most common location for accidental striking. A lower proportion of finger fractures in fall and of shoulder/forearm fractures in accidental striking should prompt the radiologist to discuss the possibility of IPV with the ED physician in any woman presenting with a finger fracture due to fall and a shoulder/forearm fracture with a vague history of accidental striking.

Upper extremity injuries in the victims of intimate partner violence

OBJECTIVES

To recognize most common patterns of upper extremity (UE) injuries in victims of Intimate Partner Violence (IPV).

METHODS

Radiological review of 308 patients who reported physical IPV at our institution from January 2013 to June 2018, identified 55 patients with 88 unique UE injuries. Demographic data and injury patterns and associations were collected from the electronic medical records.

RESULTS

The cohort included 49 females and 6 males (age 19-63, mean 38). At the time of injury, IPV was reported in 15/88 (17%) and IPV screening was documented for 22/88 (25%) injuries. There were 46 fractures, 8 dislocations or subluxations, and 34 isolated soft tissue injuries, most commonly involving the hand (56/88). Fractures most commonly involved the fingers (21/46, 46%) and the 5th digit (8/27, 30%). Medial UE fractures (5th digit, 4th digit) constituted 44% of hand and finger fractures (12/27) and 26% of all fractures (12/46). Comminuted and displaced fractures were rare (8/46, 17%). Head and face injuries were the most common concomitant injuries (9/22, 41%) and subsequent injuries (21/61, 35%). Of 12 patients with recurrent UE injuries, 6 had recurrent injuries of the same hand. Five of 6 non-acute fractures (83%) were of the hand.

CONCLUSIONS

Hand and finger injuries are the most common UE injuries in patients with IPV, with finger being the most common site and medial hand the most common region of fracture. Repeated injuries involving the same site and a combination of medial hand and head or face injuries could indicate IPV.

KEY POINTS

• Upper extremity injuries in victims of intimate partner violence are most commonly seen in the hand and fingers. • Fingers are the most common site of fracture and the medial hand is the most common region of fracture in the upper extremity in victims of intimate partner violence. • In intimate partner violence victims with upper extremity injuries, concomitant injuries and subsequent injuries are most commonly seen in the head and neck region.

Recognizing Isolated Ulnar Fractures as Potential Markers for Intimate Partner Violence

PURPOSE

This study aimed to assess the incidence of intimate partner violence (IPV) in women with isolated ulnar fractures and compare the injury characteristics in victims of IPV with those who sustained the same fractures due to other causes.

METHODS

Electronic health records from three level I trauma centers were queried to identify a cohort of women, aged 18 to 50, sustaining isolated ulnar fractures from 2005 to 2019. Radiographs were reviewed for fracture location, comminution, and displacement. Demographic data, number of visits to the emergency department, and documentation of IPV were also collected. Patients were stratified into four groups based on clinical chart review: confirmed IPV, possible IPV, not suspected for IPV, and not IPV. Historical imaging analysis for IPV prediction was also performed.

RESULTS

There were 62 patients, with a mean age of 31 years (IPV: 12 confirmed, 8 possible, 8 suspected not IPV, 34 confirmed not IPV). Comparative analysis with and without suspected cases demonstrated IPV to be associated with nondisplaced fractures (95% versus 43%; P < .001 and 91% versus 44%; P = .012). Confirmed cases were also associated with homelessness (46% versus 0%; P < .001), and the number of documented emergency department visits (median 7.0; interquartile range 2.0-12.8 versus 1.0; interquartile range 1.0-2.0; P < .001). Formal documentation of IPV evaluation was completed in only 14 of 62 (22.5%) patients. Historical imaging analysis predicted IPV in 8 of 12 (75%) confirmed IPV cases.

CONCLUSION

Up to one-third of adult women sustaining isolated ulnar fractures may be the victims of IPV. Lack of displacement on radiographs, frequent emergency department visits, and homelessness would favor IPV etiology.

Addressing intimate partner violence during the COVID-19 pandemic and beyond: how radiologists can make a difference

Faced with the COVID-19 pandemic, many countries both in Europe and across the world implemented strict stay-at-home orders. These measures helped to slow the spread of the coronavirus but also led to increased mental and physical health issues for the domestically confined population, including an increase in the occurrence of intimate partner violence (IPV) in many countries. IPV is defined as behavior that inflicts physical, psychological, or sexual harm within an intimate relationship. We believe that as radiologists, we can make a difference by being cognizant of this condition, raising an alert when appropriate and treating suspected victims with care and empathy. The aim of this Special Report is to raise awareness of IPV among radiologists and to suggest strategies by which to identify and support IPV victims. KEY POINTS: • The COVID-19 pandemic led to a marked increase in the number of intimate partner violence (IPV) cases, potentially leading to increased emergency department visits and radiological examinations. • Most IPV-related fractures affect the face, fingers, and upper trunk, and may easily be misinterpreted as routine trauma. • Radiologists should carefully review the medical history of suspicious cases, discuss the suspicion with the referring physician, and proactively engage in a private conversation with the patient, pointing to actionable resources for IPV victims.

Intimate Partner Violence: A Primer for Radiologists to Make the "Invisible" Visible

Intimate partner violence (IPV) is the physical, sexual, or emotional violence between current or former partners. It is a major public health issue that affects nearly one out of four women. Nonetheless, IPV is greatly underdiagnosed. Imaging has played a significant role in identifying cases of nonaccidental trauma in children, and similarly, it has the potential to enable the identification of injuries resulting from IPV. Radiologists have early access to the radiologic history of such victims and may be the first to diagnose IPV on the basis of the distribution and imaging appearance of the patient's currrent and past injuries. Radiologists must be familiar with the imaging findings that are suggestive of injuries resulting from IPV. Special attention should be given to cases in which there are multiple visits for injury care; coexistent fractures at different stages of healing, which may help differentiate injuries related to IPV from those caused by a stranger; and injuries in defensive locations and target areas such as the face and upper extremities. The authors provide an overview of current methods for diagnosing IPV and define the role of the radiologist in cases of IPV. They also describe a successful diagnostic imaging-based approach for helping to identify IPV, with a specific focus on the associated imaging findings and mechanisms of injuries. In addition, current needs and future perspectives for improving the diagnosis of this hidden epidemic are identified. This information is intended to raise awareness among radiologists, with the ultimate goal of improving the diagnosis of IPV and thus reducing the devastating effects on victims' lives. ^©RSNA, 2020.

Victims of family violence identified in emergency care: Comparisons of mental health and somatic diagnoses with other victims of interpersonal violence by a retrospective chart review

Family violence is a global health problem incurring significant costs to both individuals and health care systems. However, family violence as a cause of trauma and other health issues is often unidentified in patients attending emergency care. Better understanding of the risk factors associated with family violence could improve the identification and treatment of victimized patients in health care settings. Little longitudinal research exists on the mental and somatic health of family violence victims currently identified in EDs and little is known about how victims of family violence differ from other help-seeking victims of interpersonal violence. A total of 345 patients were identified as victims of interpersonal violence in a mid-size Finnish ED during the period 2011–2014. A retrospective chart review was conducted to analyze their mental and somatic health two years before and two years after identification. Victims of family violence were most likely women and they were significantly older than other victim groups. Victims of family violence also presented the most varied health symptoms both before and after identification, although differences between victim groups were not as clear as in previous studies comparing victims of family violence with non-victims. Comparison with previous data demonstrated that family violence was severely under-identified at the study site, further increasing the likelihood of family violence victims revisiting health care services. More attention should thus be paid to the identification and treatment of family violence in emergency care and other health care settings.

Disparities in Suicidality by Gender Identity Among Medicare Beneficiaries

INTRODUCTION

Suicidality is higher for gender minorities than the general population, yet little is known about suicidality in disabled or older adult gender minorities.

METHODS

This study used 2009-2014 Medicare claims to identify people with gender identity-related diagnosis codes (disabled, n=6,678; older adult, n=2,018) and compared their prevalence of suicidality with a 5% random non-gender minority beneficiary sample (disabled, n=535,801; older adult, n=1,700,008). Correlates of suicidality were assessed (via chi-square) for each of the 4 participant groups separately, and then disparities within eligibility status (disabled or older adult) were assessed using logistic regression models, adjusting first for age and mental health chronic conditions and then additionally for Medicaid eligibility, race/ethnicity, or U.S. region (each separately). The primary hypotheses were that gender minority beneficiaries would have higher suicidality but that suicidality disparities would persist after adjusting for covariates. Data were analyzed between 2017 and 2019.

RESULTS

Gender minority beneficiaries had higher unadjusted suicidality than non-gender minority beneficiaries in the disabled cohort (18.5% vs 7.1%, p<0.001). Significant suicidality predictors in all 4 groups included the following: age (except in older adult gender minorities), Medicaid eligibility, depression or behavioral health conditions, avoidable hospitalizations, and violence victimization. In age- and mental health-adjusted logistic regression models, gender minorities had higher odds of suicidality than non-gender minority beneficiaries (disabled, OR=1.95, p<0.0001; older adult, OR=2.10, p<0.0001). Disparities were not attenuated after adjusting for Medicaid eligibility, race/ethnicity, or region.

CONCLUSIONS

Heightened suicidality among identified gender minority Medicare beneficiaries highlights a pressing need to identify and reduce barriers to wellness in this population.

A Review of Challenges and Opportunities in Machine Learning for Health

Modern electronic health records (EHRs) provide data to answer clinically meaningful questions. The growing data in EHRs makes healthcare ripe for the use of machine learning. However, learning in a clinical setting presents unique challenges that complicate the use of common machine learning methodologies. For example, diseases in EHRs are poorly labeled, conditions can encompass multiple underlying endotypes, and healthy individuals are underrepresented. This article serves as a primer to illuminate these challenges and highlights opportunities for members of the machine learning community to contribute to healthcare.

Clinical features in primary care electronic records before diagnosis of ankylosing spondylitis: a nested case-control study

Background

Ankylosing spondylitis (AS) often has a long period from first symptom presentation to diagnosis. We examined the occurrence of symptoms, prescriptions and diagnostic tests in primary care electronic records over time prior to a diagnosis of AS.

Methods

Nested case-control study using anonymised primary care electronic health records from Scotland. Cases were 74 adults with a first diagnosis of AS between 2000 and 2010. Controls were matched for age, sex and GP practice: (a) 296 randomly selected adults (b) 169 adults whose records contained codes indicating spinal conditions or symptoms.

We extracted clinical features (symptoms, AS-related disorders, prescriptions and diagnostic tests). Conditional logistic regression was used to examine the association between clinical features (both individually and in combinations) and diagnosis of AS. We examined the associations between clinical features and diagnosis over time prior to diagnosis.

Results

Several new composite pointers were predictive of AS: including distinct episodes of axial pain separated by more than 6 months (OR 12.7, 95% CI 4.7 to 34.6); the occurrence of axial pain with and tendon symptoms within the same year (OR 21.7, 95% CI 2.6 to 181.5); and the co-occurrence (within 30 days) of axial pain and a prescription for nonsteroidal anti-inflammatory drug (OR 10.4, 95%CI 4.9 to 22.1). Coded episodes of axial pain increased steadily over the 3 years before diagnosis. In contrast, large joint symptoms and enthesopathy showed little or no time trend prior to diagnosis.

Conclusions

We identified novel composite pointers to a diagnosis of AS in GP records. These may represent valuable targets for diagnostic support systems.

Clinical features in primary care electronic records before diagnosis of Ankylosing Spondylitis: a nested case-control study.. [DOI: 10.21203/rs.2.17268/v4] [Citation(s) in RCA: 0] [Impact Index Per Article: 0]

Background Ankylosing spondylitis (AS) often has a long period from first symptom presentation to diagnosis. We examined the occurrence of symptoms, prescriptions and diagnostic tests in primary care electronic records over time prior to a diagnosis of AS.Methods Nested case-control study using anonymised primary care electronic health records from Scotland. Cases were 74 adults with a first diagnosis of AS between 2000 and 2010. Controls were matched for age, sex and GP practice: (a) 296 randomly selected adults (b) 169 adults whose records contained codes indicating spinal conditions or symptoms. We extracted clinical features (symptoms, AS-related disorders, prescriptions and diagnostic tests). Conditional logistic regression was used to examine the association between clinical features (both individually and in combinations) and diagnosis of AS. We examined the associations between clinical features and diagnosis over time prior to diagnosis.Results Several new composite pointers were predictive of AS: including distinct episodes of axial pain separated by more than 6 months (OR 12.7, 95% CI 4.7 to 34.6); the occurrence of axial pain with and tendon symptoms within the same year (OR 21.7, 95% CI 2.6 to 181.5); and the co-occurrence (within 30 days) of axial pain and a prescription for nonsteroidal anti-inflammatory drug (OR 10.4, 95%CI 4.9 to 22.1). Coded episodes of axial pain increased steadily over the three years before diagnosis. In contrast, large joint symptoms and enthesopathy showed little or no time trend prior to diagnosis. Conclusions We identified novel composite pointers to a diagnosis of AS in GP records. These may represent valuable targets for diagnostic support systems.

Clinical features in primary care electronic records before diagnosis of Ankylosing Spondylitis: a nested case-control study.. [DOI: 10.21203/rs.2.17268/v3] [Citation(s) in RCA: 0] [Impact Index Per Article: 0]

Background Ankylosing spondylitis (AS) often has a long period from first symptom presentation to diagnosis. We examined the occurrence of symptoms, prescriptions and diagnostic tests in primary care electronic records over time prior to a diagnosis of AS. Methods Nested case-control study using anonymised primary care electronic health records from Scotland. Cases were 74 adults with a first diagnosis of AS between 2000 and 2010. Controls were matched for age, sex and GP practice: (a) 296 randomly selected adults (b) 169 adults whose records contained codes indicating spinal conditions or symptoms. We extracted clinical features (symptoms, AS-related disorders, prescriptions and diagnostic tests). Conditional logistic regression was used to examine the association between clinical features (both individually and in combinations) and diagnosis of AS. We examined the associations between clinical features and diagnosis over time prior to diagnosis. Results Several new composite pointers were predictive of AS: including distinct episodes of axial pain separated by more than 6 months (OR 12.7, 95% CI 4.7 to 34.6); the occurrence of axial pain with and tendon symptoms within the same year (OR 21.7, 95% CI 2.6 to 181.5); and the co-occurrence (within 30 days) of axial pain and a prescription for nonsteroidal anti-inflammatory drug (OR 10.4, 95%CI 4.9 to 22.1). Coded episodes of axial pain increased steadily over the three years before diagnosis. In contrast, large joint symptoms and enthesopathy showed little or no time trend prior to diagnosis. Conclusions We identified novel composite pointers to a diagnosis of AS in GP records. These may represent valuable targets for diagnostic support systems.

Validation of an Electronic Health Record-Based Suicide Risk Prediction Modeling Approach Across Multiple Health Care Systems

Importance

Suicide is a leading cause of mortality, with suicide-related deaths increasing in recent years. Automated methods for individualized risk prediction have great potential to address this growing public health threat. To facilitate their adoption, they must first be validated across diverse health care settings.

Objective

To evaluate the generalizability and cross-site performance of a risk prediction method using readily available structured data from electronic health records in predicting incident suicide attempts across multiple, independent, US health care systems.

Design, Setting, and Participants

For this prognostic study, data were extracted from longitudinal electronic health record data comprising International Classification of Diseases, Ninth Revision diagnoses, laboratory test results, procedures codes, and medications for more than 3.7 million patients from 5 independent health care systems participating in the Accessible Research Commons for Health network. Across sites, 6 to 17 years' worth of data were available, up to 2018. Outcomes were defined by International Classification of Diseases, Ninth Revision codes reflecting incident suicide attempts (with positive predictive value >0.70 according to expert clinician medical record review). Models were trained using naive Bayes classifiers in each of the 5 systems. Models were cross-validated in independent data sets at each site, and performance metrics were calculated. Data analysis was performed from November 2017 to August 2019.

Main Outcomes and Measures

The primary outcome was suicide attempt as defined by a previously validated case definition using International Classification of Diseases, Ninth Revision codes. The accuracy and timeliness of the prediction were measured at each site.

Results

Across the 5 health care systems, of the 3 714 105 patients (2 130 454 female [57.2%]) included in the analysis, 39 162 cases (1.1%) were identified. Predictive features varied by site but, as expected, the most common predictors reflected mental health conditions (eg, borderline personality disorder, with odds ratios of 8.1-12.9, and bipolar disorder, with odds ratios of 0.9-9.1) and substance use disorders (eg, drug withdrawal syndrome, with odds ratios of 7.0-12.9). Despite variation in geographical location, demographic characteristics, and population health characteristics, model performance was similar across sites, with areas under the curve ranging from 0.71 (95% CI, 0.70-0.72) to 0.76 (95% CI, 0.75-0.77). Across sites, at a specificity of 90%, the models detected a mean of 38% of cases a mean of 2.1 years in advance.

Conclusions and Relevance

Across 5 diverse health care systems, a computationally efficient approach leveraging the full spectrum of structured electronic health record data was able to detect the risk of suicidal behavior in unselected patients. This approach could facilitate the development of clinical decision support tools that inform risk reduction interventions.

Clinical features in primary care electronic records before diagnosis of Ankylosing Spondylitis: a nested case-control study.. [DOI: 10.21203/rs.2.17268/v2] [Citation(s) in RCA: 0] [Impact Index Per Article: 0]

Background Ankylosing spondylitis (AS) often has a long period from first symptom presentation to diagnosis. We examined the occurrence of symptoms, prescriptions and diagnostic tests in primary care electronic records over time prior to a diagnosis of AS.Methods Nested case-control study using anonymised primary care electronic health records from Scotland. Cases were 100 adults with a first diagnosis of AS between 1994 and 2010. Controls were matched for age, sex and GP practice: (a) 400 randomly selected adults (b) 236 adults whose records contained codes indicating spinal conditions or symptoms. We extracted clinical features (symptoms, AS-related disorders, prescriptions and diagnostic tests). Conditional logistic regression was used to examine the association between clinical features (both individually and in combinations) and diagnosis of AS. We examined the associations between clinical features and diagnosis over time prior to diagnosis.Results Several new composite pointers were predictive of AS: including distinct episodes of axial pain separated by more than 6 months (OR 3.7, 95% CI 1.8 to 7.5) and the co-occurrence of axial pain with either large joint symptoms (OR 2.7, 95% CI 1.1 to 6.3) or tendon symptoms within the same year (OR 3.6, 95% CI 1.3 to 10.3). Coded episodes of axial pain increased steadily over the three years before diagnosis. In contrast, large joint symptoms and enthesopathy showed little or no time trend prior to diagnosis.Conclusions We identified novel composite pointers to a diagnosis of AS in GP records. These may represent valuable targets for diagnostic support systems.

Development of an individualized risk calculator for poor functioning in young people victimized during childhood: A longitudinal cohort study

BACKGROUND

Childhood victimization elevates the average risk of developing functional impairment in adulthood. However, not all victimized children demonstrate poor outcomes. Although research has described factors that confer vulnerability or resilience, it is unknown if this knowledge can be translated to accurately identify the most vulnerable victimized children.

OBJECTIVE

To build and internally validate a risk calculator to identify those victimized children who are most at risk of functional impairment at age 18 years.

PARTICIPANTS

We utilized data from the Environmental Risk (E-Risk) Longitudinal Twin Study, a nationally-representative birth cohort of 2232 UK children born in 1994-95.

METHODS

Victimization exposure was assessed repeatedly between ages 5 and 12 years along with a range of individual-, family- and community-level predictors. Functional outcomes were assessed at age 18 years. We developed and evaluated a prediction model for psychosocial disadvantage and economic disadvantage using the Least Absolute Shrinkage and Selection Operator (LASSO) regularized regression with nested 10-fold cross-validation.

RESULTS

The model predicting psychosocial disadvantage following childhood victimization retained 12 of 22 predictors, had an area under the curve (AUC) of 0.65, and was well-calibrated within the range of 40-70% predicted risk. The model predicting economic disadvantage retained 10 of 22 predictors, achieved excellent discrimination (AUC = 0.80), and a high degree of calibration.

CONCLUSIONS

Prediction modelling techniques can be applied to estimate individual risk for poor functional outcomes in young adulthood following childhood victimization. Such risk prediction tools could potentially assist practitioners to target interventions, which is particularly useful in a context of scarce resources.

Predicting mechanical restraint of psychiatric inpatients by applying machine learning on electronic health data

OBJECTIVE

Mechanical restraint (MR) is used to prevent patients from harming themselves or others during inpatient treatment. The objective of this study was to investigate whether incident MR occurring in the first 3 days following admission could be predicted based on analysis of electronic health data available after the first hour of admission.

METHODS

The dataset consisted of clinical notes from electronic health records from the Central Denmark Region and data from the Danish Health Registers from patients admitted to a psychiatric department in the period from 2011 to 2015. Supervised machine learning algorithms were trained on a randomly selected subset of the data and validated using an independent test dataset.

RESULTS

A total of 5050 patients with 8869 admissions were included in the study. One hundred patients were mechanically restrained in the period between one hour and 3 days after the admission. A Random Forest algorithm predicted MR with an area under the curve of 0.87 (95% CI 0.79-0.93). At 94% specificity, the sensitivity was 56%. Among the ten strongest predictors, nine were derived from the clinical notes.

CONCLUSIONS

These findings open for the development of an early warning system that may guide interventions to reduce the use of MR.

The use of electronic health records for psychiatric phenotyping and genomics

The widespread adoption of electronic health record (EHRs) in healthcare systems has created a vast and continuously growing resource of clinical data and provides new opportunities for population-based research. In particular, the linking of EHRs to biospecimens and genomic data in biobanks may help address what has become a rate-limiting study for genetic research: the need for large sample sizes. The principal roadblock to capitalizing on these resources is the need to establish the validity of phenotypes extracted from the EHR. For psychiatric genetic research, this represents a particular challenge given that diagnosis is based on patient reports and clinician observations that may not be well-captured in billing codes or narrative records. This review addresses the opportunities and pitfalls in EHR-based phenotyping with a focus on their application to psychiatric genetic research. A growing number of studies have demonstrated that diagnostic algorithms with high positive predictive value can be derived from EHRs, especially when structured data are supplemented by text mining approaches. Such algorithms enable semi-automated phenotyping for large-scale case-control studies. In addition, the scale and scope of EHR databases have been used successfully to identify phenotypic subgroups and derive algorithms for longitudinal risk prediction. EHR-based genomics are particularly well-suited to rapid look-up replication of putative risk genes, studies of pleiotropy (phenomewide association studies or PheWAS), investigations of genetic networks and overlap across the phenome, and pharmacogenomic research. EHR phenotyping has been relatively under-utilized in psychiatric genomic research but may become a key component of efforts to advance precision psychiatry.

Pointers to earlier diagnosis of endometriosis: a nested case-control study using primary care electronic health records

BACKGROUND

Endometriosis is a condition with relatively non-specific symptoms, and in some cases a long time elapses from first-symptom presentation to diagnosis.

AIM

To develop and test new composite pointers to a diagnosis of endometriosis in primary care electronic records.

DESIGN AND SETTING

This is a nested case-control study of 366 cases using the Practice Team Information database of anonymised primary care electronic health records from Scotland. Data were analysed from 366 cases of endometriosis between 1994 and 2010, and two sets of age and GP practice matched controls: (a) 1453 randomly selected females and (b) 610 females whose records contained codes indicating consultation for gynaecological symptoms.

METHOD

Composite pointers comprised patterns of symptoms, prescribing, or investigations, in combination or over time. Conditional logistic regression was used to examine the presence of both new and established pointers during the 3 years before diagnosis of endometriosis and to identify time of appearance.

RESULTS

A number of composite pointers that were strongly predictive of endometriosis were observed. These included pain and menstrual symptoms occurring within the same year (odds ratio [OR] 6.5, 95% confidence interval [CI] = 3.9 to 10.6), and lower gastrointestinal symptoms occurring within 90 days of gynaecological pain (OR 6.1, 95% CI = 3.6 to 10.6). Although the association of infertility with endometriosis was only detectable in the year before diagnosis, several pain-related features were associated with endometriosis several years earlier.

CONCLUSION

Useful composite pointers to a diagnosis of endometriosis in GP records were identified. Some of these were present several years before the diagnosis and may be valuable targets for diagnostic support systems.

Grappling with the Future Use of Big Data for Translational Medicine and Clinical Care

Objectives: Although patients may have a wealth of imaging, genomic, monitoring, and personal device data, it has yet to be fully integrated into clinical care. Methods: We identify three reasons for the lack of integration. The first is that "Big Data" is poorly managed by most Electronic Medical Record Systems (EMRS). The data is mostly available on "cloud-native" platforms that are outside the scope of most EMRs, and even checking if such data is available on a patient often must be done outside the EMRS. The second reason is that extracting features from the Big Data that are relevant to healthcare often requires complex machine learning algorithms, such as determining if a genomic variant is protein-altering. The third reason is that applications that present Big Data need to be modified constantly to reflect the current state of knowledge, such as instructing when to order a new set of genomic tests. In some cases, applications need to be updated nightly. Results: A new architecture for EMRS is evolving which could unite Big Data, machine learning, and clinical care through a microservice-based architecture which can host applications focused on quite specific aspects of clinical care, such as managing cancer immunotherapy. Conclusion: Informatics innovation, medical research, and clinical care go hand in hand as we look to infuse science-based practice into healthcare. Innovative methods will lead to a new ecosystem of applications (Apps) interacting with healthcare providers to fulfill a promise that is still to be determined.

Predicting early psychiatric readmission with natural language processing of narrative discharge summaries

The ability to predict psychiatric readmission would facilitate the development of interventions to reduce this risk, a major driver of psychiatric health-care costs. The symptoms or characteristics of illness course necessary to develop reliable predictors are not available in coded billing data, but may be present in narrative electronic health record (EHR) discharge summaries. We identified a cohort of individuals admitted to a psychiatric inpatient unit between 1994 and 2012 with a principal diagnosis of major depressive disorder, and extracted inpatient psychiatric discharge narrative notes. Using these data, we trained a 75-topic Latent Dirichlet Allocation (LDA) model, a form of natural language processing, which identifies groups of words associated with topics discussed in a document collection. The cohort was randomly split to derive a training (70%) and testing (30%) data set, and we trained separate support vector machine models for baseline clinical features alone, baseline features plus common individual words and the above plus topics identified from the 75-topic LDA model. Of 4687 patients with inpatient discharge summaries, 470 were readmitted within 30 days. The 75-topic LDA model included topics linked to psychiatric symptoms (suicide, severe depression, anxiety, trauma, eating/weight and panic) and major depressive disorder comorbidities (infection, postpartum, brain tumor, diarrhea and pulmonary disease). By including LDA topics, prediction of readmission, as measured by area under receiver-operating characteristic curves in the testing data set, was improved from baseline (area under the curve 0.618) to baseline+1000 words (0.682) to baseline+75 topics (0.784). Inclusion of topics derived from narrative notes allows more accurate discrimination of individuals at high risk for psychiatric readmission in this cohort. Topic modeling and related approaches offer the potential to improve prediction using EHRs, if generalizability can be established in other clinical cohorts.

Laboratory tests as short-term correlates of stroke

BACKGROUND

The widespread adoption of electronic health records provides new opportunities to better predict which patients are likely to suffer a stroke. Using electronic health records, we assessed the correlation of different laboratory tests to future occurrences of a stroke.

METHODS

We examined the electronic health records of 2.4 million people over a two year time span. These records contained 26,964 diagnoses of stroke. Using Cox regression analysis, we measured whether any one of 1796 different laboratory tests were effectively correlated with a future diagnosis of stroke.

RESULTS

We identified 38 different laboratory tests that had significant short-term (two year) prognostic value for a future diagnosis of stroke. For each of the 38 laboratory tests we also compiled the Kaplan-Meier survival curve, and relative risk ratio that the test confers.

CONCLUSION

Several dozen laboratory tests are effective short-term correlates of stroke.

Integrating intimate partner violence assessment and intervention into healthcare in the United States: a systems approach

The Institute of Medicine, United States Preventive Services Task Force (USPSTF), and national healthcare organizations recommend screening and counseling for intimate partner violence (IPV) within the US healthcare setting. The Affordable Care Act includes screening and brief counseling for IPV as part of required free preventive services for women. Thus, IPV screening and counseling must be implemented safely and effectively throughout the healthcare delivery system. Health professional education is one strategy for increasing screening and counseling in healthcare settings, but studies on improving screening and counseling for other health conditions highlight the critical role of making changes within the healthcare delivery system to drive desired improvements in clinician screening practices and health outcomes. This article outlines a systems approach to the implementation of IPV screening and counseling, with a focus on integrated health and advocacy service delivery to support identification and interventions, use of electronic health record (EHR) tools, and cross-sector partnerships. Practice and policy recommendations include (1) ensuring staff and clinician training in effective, client-centered IPV assessment that connects patients to support and services regardless of disclosure; (2) supporting enhancement of EHRs to prompt appropriate clinical care for IPV and facilitate capturing more detailed and standardized IPV data; and (3) integrating IPV care into quality and meaningful use measures. Research directions include studies across various health settings and populations, development of quality measures and patient-centered outcomes, and tests of multilevel approaches to improve the uptake and consistent implementation of evidence-informed IPV screening and counseling guidelines.

Using Electronic Health Records for Population Health Research: A Review of Methods and Applications

The use and functionality of electronic health records (EHRs) have increased rapidly in the past decade. Although the primary purpose of EHRs is clinical, researchers have used them to conduct epidemiologic investigations, ranging from cross-sectional studies within a given hospital to longitudinal studies on geographically distributed patients. Herein, we describe EHRs, examine their use in population health research, and compare them with traditional epidemiologic methods. We describe diverse research applications that benefit from the large sample sizes and generalizable patient populations afforded by EHRs. These have included reevaluation of prior findings, a range of diseases and subgroups, environmental and social epidemiology, stigmatized conditions, predictive modeling, and evaluation of natural experiments. Although studies using primary data collection methods may have more reliable data and better population retention, EHR-based studies are less expensive and require less time to complete. Future EHR epidemiology with enhanced collection of social/behavior measures, linkage with vital records, and integration of emerging technologies such as personal sensing could improve clinical care and population health.

Area disease estimation based on sentinel hospital records

BACKGROUND

Population health attributes (such as disease incidence and prevalence) are often estimated using sentinel hospital records, which are subject to multiple sources of uncertainty. When applied to these health attributes, commonly used biased estimation techniques can lead to false conclusions and ineffective disease intervention and control. Although some estimators can account for measurement error (in the form of white noise, usually after de-trending), most mainstream health statistics techniques cannot generate unbiased and minimum error variance estimates when the available data are biased.

METHODS AND FINDINGS

A new technique, called the Biased Sample Hospital-based Area Disease Estimation (B-SHADE), is introduced that generates space-time population disease estimates using biased hospital records. The effectiveness of the technique is empirically evaluated in terms of hospital records of disease incidence (for hand-foot-mouth disease and fever syndrome cases) in Shanghai (China) during a two-year period. The B-SHADE technique uses a weighted summation of sentinel hospital records to derive unbiased and minimum error variance estimates of area incidence. The calculation of these weights is the outcome of a process that combines: the available space-time information; a rigorous assessment of both, the horizontal relationships between hospital records and the vertical links between each hospital's records and the overall disease situation in the region. In this way, the representativeness of the sentinel hospital records was improved, the possible biases of these records were corrected, and the generated area incidence estimates were best linear unbiased estimates (BLUE). Using the same hospital records, the performance of the B-SHADE technique was compared against two mainstream estimators.

CONCLUSIONS

The B-SHADE technique involves a hospital network-based model that blends the optimal estimation features of the Block Kriging method and the sample bias correction efficiency of the ratio estimator method. In this way, B-SHADE can overcome the limitations of both methods: Block Kriging's inadequacy concerning the correction of sample bias and spatial clustering; and the ratio estimator's limitation as regards error minimization. The generality of the B-SHADE technique is further demonstrated by the fact that it reduces to Block Kriging in the case of unbiased samples; to ratio estimator if there is no correlation between hospitals; and to simple statistic if the hospital records are neither biased nor space-time correlated. In addition to the theoretical advantages of the B-SHADE technique over the two other methods above, two real world case studies (hand-foot-mouth disease and fever syndrome cases) demonstrated its empirical superiority, as well.

A predictive model to help identify intimate partner violence based on diagnoses and phone calls

BACKGROUND

Intimate partner violence (IPV) is a significant health problem but goes largely undiagnosed, undisclosed, and clinically undocumented.

PURPOSE

To use historical data on diagnoses and telephone advice calls to develop a predictive model that identifies clinical profiles of women at high risk for undisclosed IPV.

METHODS

A case-control study was conducted in women aged 18-44 years enrolled at Kaiser Permanente Northern California (KPNC) in 2005-2006 using symptoms reported by telephone and clinical diagnosis from electronic medical records. Analysis was conducted in 2007-2010. Overall, 1276 cases were identified using ICD-9 codes for IPV and were matched with 5 controls each. A full multivariate model was developed to identify those with IPV, as well as a reduced model and a summed-score model whose performance characteristics were assessed.

RESULTS

Predictors most highly associated with IPV were history of remote IPV (OR=7.8); calls or diagnoses for psychiatric problems (OR=2.4); calls for HIV concerns (OR=2.4); and clinical diagnoses of prenatal complications (OR=2.1). Using the summed-score model for a population with IPV prevalence of 7%, and using a threshold score of 3 for predicting IPV with a sensitivity of 75%, 9.7 women would need to be assessed to diagnose one case of IPV.

CONCLUSIONS

Diagnosed IPV was associated with a clinical profile based on both telephone call data and clinical diagnoses. The simple predictive model can prompt focused clinical inquiry and improve diagnosis of IPV in any clinical setting.