Regression tree boosting to adjust health care cost predictions for diagnostic mix

Using Data-Driven Approaches to Classify and Predict Health Care Spending in Patients With Gout Using Urate-Lowering Therapy

OBJECTIVE

Despite increasing overall health care spending over the past several decades, little is known about long-term patterns of spending among US patients with gout. Current approaches to assessing spending typically focus on composite measures or patients agnostic to disease state; in contrast, examining spending using longitudinal measures may better discriminate patients and target interventions to those in need. We used a data-driven approach to classify and predict spending patterns in patients with gout.

METHODS

Using insurance claims data from 2017-2019, we used group-based trajectory modeling to classify patients ages 40 years or older diagnosed with gout and treated with urate-lowering therapy (ULT) by their total health care spending over 2 years. We assessed the ability to predict membership in each spending group using logistic and generalized boosted regression with split-sample validation. Models were estimated using different sets of predictors and evaluated using C statistics.

RESULTS

In 57,980 patients, the mean ± SD age was 71.0 ± 10.5 years, and 17,194 patients (29.7%) were female. The best-fitting model included the following groups: minimal spending (13.2%), moderate spending (37.4%), and high spending (49.4%). The ability to predict groups was high overall (e.g., boosted C statistics with all predictors: minimal spending [0.89], moderate spending [0.78], and high spending [0.90]). Although average adherence was relatively high in the population, for the high-spending group, the most influential predictors were greater gout medication adherence and diabetes melllitus diagnosis.

CONCLUSION

We identified distinct long-term health care spending patterns in patients with gout using ULT with high accuracy. Several clinical predictors could be key areas for intervention, such as gout medication use or diabetes melllitus.

Combining Resampling Strategies and Ensemble Machine Learning Methods to Enhance Prediction of Neonates with a Low Apgar Score After Induction of Labor in Northern Tanzania

Objective

The goal of this study was to establish the most efficient boosting method in predicting neonatal low Apgar scores following labor induction intervention and to assess whether resampling strategies would improve the predictive performance of the selected boosting algorithms.

Methods

A total of 7716 singleton births delivered from 2000 to 2015 were analyzed. Cesarean deliveries following labor induction, deliveries with abnormal presentation, and deliveries with missing Apgar score or delivery mode information were excluded. We examined the effect of resampling approaches or data preprocessing on predicting low Apgar scores, specifically the synthetic minority oversampling technique (SMOTE), borderline-SMOTE, and the random undersampling (RUS) technique. Sensitivity, specificity, precision, area under receiver operating curve (AUROC), F-score, positive predicted values (PPV), negative predicted values (NPV) and accuracy of the three (3) boosting-based ensemble methods were used to evaluate their discriminative ability. The ensemble learning models tested include adoptive boosting (AdaBoost), gradient boosting (GB) and extreme gradient boosting method (XGBoost).

Results

The prevalence of low (<7) Apgar scores was 9.5% (n = 733). The prediction models performed nearly similar in their baseline mode. Following the application of resampling techniques, borderline-SMOTE significantly improved the predictive performance of all the boosting-based ensemble methods under observation in terms of sensitivity, F1-score, AUROC and PPV.

Conclusion

Policymakers, healthcare informaticians and neonatologists should consider implementing data preprocessing strategies when predicting a neonatal outcome with imbalanced data to enhance efficiency. The process may be more effective when borderline-SMOTE technique is deployed on the selected ensemble classifiers. However, future research may focus on testing additional resampling techniques, performing feature engineering, variable selection and optimizing further the ensemble learning hyperparameters.

Association of Breast Implants with Nonspecific Symptoms, Connective Tissue Diseases, and Allergic Reactions: A Retrospective Cohort Analysis

BACKGROUND

Given the rising media attention regarding various adverse conditions attributed to breast implants, the authors examined the association between breast implantation and the risk of being diagnosed with connective tissue diseases, allergic reactions, and nonspecific constitutional complaints in a cohort study with longitudinal follow-up.

METHODS

Women enrolled in a regional military health care system between 2003 and 2012 were evaluated in this retrospective cohort study. A propensity score was generated to match women who underwent breast implantation with women who did not undergo breast implantation. The propensity score included age, social history, health care use, comorbidities, and medication use. Outcomes assessed included International Classification of Diseases, Ninth Revision, diagnoses codes for (1) nonspecific constitutional symptoms, (2) nonspecific cardiac conditions, (3) rheumatoid arthritis and systemic lupus erythematosus, (4) other connective tissue diseases, and (5) allergic reactions.

RESULTS

Of 22,063 women included in the study (513 breast implants and 21,550 controls), we propensity score-matched 452 breast implant recipients with 452 nonrecipients. Odds ratios and 95 percent confidence intervals in breast implant recipients compared to nonrecipients were similar, including nonspecific constitutional symptoms (OR, 0.77; 95 percent CI, 0.53 to 1.13), nonspecific cardiac conditions (OR, 0.97; 95 percent CI, 0.69 to 1.37), rheumatoid arthritis and systemic lupus erythematosus (OR, 0.66; 95 percent CI, 0.33 to 1.31), other connective tissue diseases (OR, 1.02; 95 percent CI, 0.78 to 1.32), and allergic reactions (OR, 1.18; 95 percent CI, 0.84 to 1.66).

CONCLUSIONS

Women with breast implants did not have an increased likelihood of being diagnosed with nonspecific constitutional symptoms, connective tissue disorders, and/or allergic reaction conditions.

CLINICAL QUESTION/LEVEL OF EVIDENCE

Therapeutic, III.

A Brief Analysis of Key Machine Learning Methods for Predicting Medicare Payments Related to Physical Therapy Practices in the United States

Background and objectives: Machine learning approaches using random forest have been effectively used to provide decision support in health and medical informatics. This is especially true when predicting variables associated with Medicare reimbursements. However, more work is needed to analyze and predict data associated with reimbursements through Medicare and Medicaid services for physical therapy practices in the United States. The key objective of this study is to analyze different machine learning models to predict key variables associated with Medicare standardized payments for physical therapy practices in the United States. Materials and Methods: This study employs five methods, namely, multiple linear regression, decision tree regression, random forest regression, K-nearest neighbors, and linear generalized additive model, (GAM) to predict key variables associated with Medicare payments for physical therapy practices in the United States. Results: The study described in this article adds to the body of knowledge on the effective use of random forest regression and linear generalized additive model in predicting Medicare Standardized payment. It turns out that random forest regression may have any edge over other methods employed for this purpose. Conclusions: The study provides a useful insight into comparing the performance of the aforementioned methods, while identifying a few intricate details associated with predicting Medicare costs while also ascertaining that linear generalized additive model and random forest regression as the most suitable machine learning models for predicting key variables associated with standardized Medicare payments.

Intersections of machine learning and epidemiological methods for health services research

The field of health services research is broad and seeks to answer questions about the health care system. It is inherently interdisciplinary, and epidemiologists have made crucial contributions. Parametric regression techniques remain standard practice in health services research with machine learning techniques currently having low penetrance in comparison. However, studies in several prominent areas, including health care spending, outcomes and quality, have begun deploying machine learning tools for these applications. Nevertheless, major advances in epidemiological methods are also as yet underleveraged in health services research. This article summarizes the current state of machine learning in key areas of health services research, and discusses important future directions at the intersection of machine learning and epidemiological methods for health services research.

Use of Data-Driven Methods to Predict Long-term Patterns of Health Care Spending for Medicare Patients

Importance

Current approaches to predicting health care costs generally rely on a single composite value of spending and focus on short time horizons. By contrast, examining patients' spending patterns using dynamic measures applied over longer periods may better identify patients with different spending and help target interventions to those with the greatest need.

Objective

To classify patients by their long-term, dynamic health care spending patterns using a data-driven approach and assess the ability to predict spending patterns, particularly using characteristics that are potentially modifiable through intervention.

Design, Setting, and Participants

This cohort study used a retrospective cohort design from a random nationwide sample of Medicare fee-for-service administrative claims data to identify beneficiaries aged 65 years or older with continuous eligibility from 2011 to 2013. Statistical analysis was performed from August 2018 to December 2019.

Main Outcomes and Measures

Group-based trajectory modeling was applied to the claims data to classify the Medicare beneficiaries by their total health care spending patterns over a 2-year period. The ability to predict membership in each trajectory spending group was assessed using generalized boosted regression, a data mining approach to model building and prediction, with split-sample validation. Models were estimated using (1) prior-year predictors and (2) prior-year predictors potentially modifiable through intervention measured in the claims data. These models were evaluated using validated C-statistics. The relative influence of individual predictors in the models was evaluated.

Results

Among the 329 476 beneficiaries, the mean (SD) age was 76.0 (7.2) years and 190 346 (57.8%) were female. This final 5-group model included a minimal-user group (group 1, 37 572 individuals [11.4%]), a low-cost group (group 2, 48 575 individuals [14.7%]), a rising-cost group (group 3, 24 736 individuals [7.5%]), a moderate-cost group (group 4, 83 338 individuals [25.3%]), and a high-cost group (group 5, 135 255 individuals [41.2%]). Potentially modifiable characteristics strongly predicted these patterns (C-statistics range: 0.68-0.94). For groups with progressively increasing spending in particular, the most influential factors were number of medications (relative influence: 29.2), number of office visits (relative influence: 30.3), and mean medication adherence (relative influence: 33.6).

Conclusions and Relevance

Using a data-driven approach, distinct spending patterns were identified with high accuracy. The potentially modifiable predictors of membership in the rising-cost group represent important levers for early interventions that may prevent later spending increases. This approach could be adapted by organizations to target quality improvement interventions, particularly because numerous health care organizations are increasingly using these routinely collected data.

Not there yet: using data-driven methods to predict who becomes costly among low-cost patients with type 2 diabetes

BACKGROUND

Diabetes is a leading cause of Medicare spending; predicting which individuals are likely to be costly is essential for targeting interventions. Current approaches generally focus on composite measures, short time-horizons, or patients who are already high utilizers, whose costs may be harder to modify. Thus, we used data-driven methods to classify unique clusters in Medicare claims who were initially low utilizers by their diabetes spending patterns in subsequent years and used machine learning to predict these patterns.

METHODS

We identified beneficiaries with type 2 diabetes whose spending was in the bottom 90% of diabetes care spending in a one-year baseline period in Medicare fee-for-service data. We used group-based trajectory modeling to classify unique clusters of patients by diabetes-related spending patterns over a two-year follow-up. Prediction models were estimated with generalized boosted regression, a machine learning method, using sets of all baseline predictors, diabetes predictors, and predictors that are potentially-modifiable through interventions. Each model was evaluated through C-statistics and 5-fold cross-validation.

RESULTS

Among 33,789 beneficiaries (baseline median diabetes spending: $4153), we identified 5 distinct spending patterns that could largely be predicted; of these, 68.1% of patients had consistent spending, 25.3% had spending that rose quickly, and 6.6% of patients had spending that rose progressively. The ability to predict these groups was moderate (validated C-statistics: 0.63 to 0.87). The most influential factors for those with progressively rising spending were age, generosity of coverage, prior spending, and medication adherence.

CONCLUSIONS

Patients with type 2 diabetes who were initially low spenders exhibit distinct subsequent long-term patterns of diabetes spending; membership in these patterns can be largely predicted with data-driven methods. These findings as well as applications of the overall approach could potentially inform the design and timing of diabetes or cost-containment interventions, such as medication adherence or interventions that enhance access to care, among patients with type 2 diabetes.

The Evaluation of Clinical Classifications Software Using the National Inpatient Sample Database

The Clinical Classifications Software (CCS), by grouping International Classification of Diseases (ICD), provides the capacity to better account for clinical conditions for payers, policy makers, and researchers to analyze outcomes, costs, and utilization. There is a critical need for additional research on application of CCS categories to validate the clinical condition representation and to prevent gaps in research. This study compared the event frequency and ICD codes of CCS categories with significant changes from the first three quarters of 2015 to 2016 using National Inpatient Sample data. A total of 63 of the 285 diagnostics CCS were identified with greater than 20% change, of which 32 had increased and 31 decreased over time. Due to the complexity associated with the transition from ICD-9 to ICD-10, more studies are needed to identify the reason for the changes to improve CCS use for ICD-10 and its comparability with ICD-9 based data.

AI in Healthcare: Time-Series Forecasting Using Statistical, Neural, and Ensemble Architectures

Both statistical and neural methods have been proposed in the literature to predict healthcare expenditures. However, less attention has been given to comparing predictions from both these methods as well as ensemble approaches in the healthcare domain. The primary objective of this paper was to evaluate different statistical, neural, and ensemble techniques in their ability to predict patients' weekly average expenditures on certain pain medications. Two statistical models, persistence (baseline) and autoregressive integrated moving average (ARIMA), a multilayer perceptron (MLP) model, a long short-term memory (LSTM) model, and an ensemble model combining predictions of the ARIMA, MLP, and LSTM models were calibrated to predict the expenditures on two different pain medications. In the MLP and LSTM models, we compared the influence of shuffling of training data and dropout of certain nodes in MLPs and nodes and recurrent connections in LSTMs in layers during training. Results revealed that the ensemble model outperformed the persistence, ARIMA, MLP, and LSTM models across both pain medications. In general, not shuffling the training data and adding the dropout helped the MLP models and shuffling the training data and not adding the dropout helped the LSTM models across both medications. We highlight the implications of using statistical, neural, and ensemble methods for time-series forecasting of outcomes in the healthcare domain.

Length-of-Stay Prediction for Pediatric Patients With Respiratory Diseases Using Decision Tree Methods

Accurate prediction of a patient's length-of-stay (LOS) in the hospital enables an efficient and effective management of hospital beds. This paper studies LOS prediction for pediatric patients with respiratory diseases using three decision tree methods: Bagging, Adaboost, and Random forest. A data set of 11,206 records retrieved from the hospital information system is used for analysis after preprocessing and transformation through a computation and an expansion method. Two tests, namely bisection test and periodic test, are designed to assess the performance of the prediction methods. Bagging shows the best result on the bisection test (0.296 RMSE, 0.831 R², and 0.723 Acc ± 1) for the testing set of the whole data test. The performances of the three methods are similar on the periodic test, whereas Adaboost performs slightly better than the other two methods. Results indicate that the three methods are all effective for the LOS prediction. This study also investigates the importance of different data fields to the LOS prediction, and finds that hospital treatment-related data fields contribute more to the LOS prediction than other categories of fields.

Association between meteorological variations and activities of influenza A and B across different climate zones: a multi-region modelling analysis across the globe

OBJECTIVE

To elucidate the effects of meteorological variations on the activity of influenza A and B in 11 sites across different climate regions.

METHODS

Daily numbers of laboratory-confirmed influenza A and B cases from 2011-2015 were collected from study sites where the corresponding daily mean temperature, relative humidity, wind speed and daily precipitation amount were used for boosted regression trees analysis on the marginal associations and the interaction effects.

RESULTS

Cold temperature was a major determinant that favored both influenza A and B in temperate and subtropical sites. Temperature-to-influenza A, but not influenza B, exhibited a U-shape association in subtropical and tropical sites. High relative humidity was also associated with influenza activities but was less consistent with influenza B activity. Compared with relative humidity, absolute humidity had a stronger association - it was negatively associated with influenza B activity in temperate zones, but was positively associated with both influenza A and B in subtropical and tropical zones.

CONCLUSION

The association between meteorological factors and with influenza activity is virus type specific and climate dependent. The heavy influence of temperature on influenza activity across climate zones implies that global warming is likely to have an impact on the influenza burden.

Risk prediction and segmentation models used in the United States for assessing risk in whole populations: a critical literature review with implications for nurses' role in population health management

Objective

We sought to assess the current state of risk prediction and segmentation models (RPSM) that focus on whole populations.

Materials

Academic literature databases (ie MEDLINE, Embase, Cochrane Library, PROSPERO, and CINAHL), environmental scan, and Google search engine.

Methods

We conducted a critical review of the literature focused on RPSMs predicting hospitalizations, emergency department visits, or health care costs.

Results

We identified 35 distinct RPSMs among 37 different journal articles (n = 31), websites (n = 4), and abstracts (n = 2). Most RPSMs (57%) defined their population as health plan enrollees while fewer RPSMs (26%) included an age-defined population (26%) and/or geographic boundary (26%). Most RPSMs (51%) focused on predicting hospital admissions, followed by costs (43%) and emergency department visits (31%), with some models predicting more than one outcome. The most common predictors were age, gender, and diagnostic codes included in 82%, 77%, and 69% of models, respectively.

Discussion

Our critical review of existing RPSMs has identified a lack of comprehensive models that integrate data from multiple sources for application to whole populations. Highly depending on diagnostic codes to define high-risk populations overlooks the functional, social, and behavioral factors that are of great significance to health.

Conclusion

More emphasis on including nonbilling data and providing holistic perspectives of individuals is needed in RPSMs. Nursing-generated data could be beneficial in addressing this gap, as they are structured, frequently generated, and tend to focus on key health status elements like functional status and social/behavioral determinants of health.

Alternative evaluation metrics for risk adjustment methods

Risk adjustment is instituted to counter risk selection by accurately equating payments with expected expenditures. Traditional risk-adjustment methods are designed to estimate accurate payments at the group level. However, this generates residual risks at the individual level, especially for high-expenditure individuals, thereby inducing health plans to avoid those with high residual risks. To identify an optimal risk-adjustment method, we perform a comprehensive comparison of prediction accuracies at the group level, at the tail distributions, and at the individual level across 19 estimators: 9 parametric regression, 7 machine learning, and 3 distributional estimators. Using the 2013-2014 MarketScan database, we find that no one estimator performs best in all prediction accuracies. Generally, machine learning and distribution-based estimators achieve higher group-level prediction accuracy than parametric regression estimators. However, parametric regression estimators show higher tail distribution prediction accuracy and individual-level prediction accuracy, especially at the tails of the distribution. This suggests that there is a trade-off in selecting an appropriate risk-adjustment method between estimating accurate payments at the group level and lower residual risks at the individual level. Our results indicate that an optimal method cannot be determined solely on the basis of statistical metrics but rather needs to account for simulating plans' risk selective behaviors.

A Systematic Review on Healthcare Analytics: Application and Theoretical Perspective of Data Mining

The growing healthcare industry is generating a large volume of useful data on patient demographics, treatment plans, payment, and insurance coverage—attracting the attention of clinicians and scientists alike. In recent years, a number of peer-reviewed articles have addressed different dimensions of data mining application in healthcare. However, the lack of a comprehensive and systematic narrative motivated us to construct a literature review on this topic. In this paper, we present a review of the literature on healthcare analytics using data mining and big data. Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, we conducted a database search between 2005 and 2016. Critical elements of the selected studies—healthcare sub-areas, data mining techniques, types of analytics, data, and data sources—were extracted to provide a systematic view of development in this field and possible future directions. We found that the existing literature mostly examines analytics in clinical and administrative decision-making. Use of human-generated data is predominant considering the wide adoption of Electronic Medical Record in clinical care. However, analytics based on website and social media data has been increasing in recent years. Lack of prescriptive analytics in practice and integration of domain expert knowledge in the decision-making process emphasizes the necessity of future research.

The relative benefits of claims and electronic health record data for predicting medication adherence trajectory

BACKGROUND

Healthcare providers are increasingly encouraged to improve their patients' adherence to chronic disease medications. Prediction of adherence can identify patients in need of intervention, but most prediction efforts have focused on claims data, which may be unavailable to providers. Electronic health records (EHR) are readily available and may provide richer information with which to predict adherence than is currently available through claims.

METHODS

In a linked database of complete Medicare Advantage claims and comprehensive EHR from a multi-specialty outpatient practice, we identified patients who filled a prescription for a statin, antihypertensive, or oral antidiabetic during 2011 to 2012. We followed patients to identify subsequent medication filling patterns and used group-based trajectory models to assign patients to adherence trajectories. We then identified potential predictors from both claims and EHR data and fit a series of models to evaluate the accuracy of each data source in predicting medication adherence.

RESULTS

Claims were highly predictive of patients in the worst adherence trajectory (C=0.78), but EHR data also provided good predictions (C=0.72). Among claims predictors, presence of a prior gap in filling of at least 6 days was by far the most influential predictor. In contrast, good predictions from EHR data required complex models with many variables.

CONCLUSION

EHR data can provide good predictions of adherence trajectory and therefore may be useful for providers seeking to deploy resource-intensive interventions. However, prior adherence information derived from claims is most predictive, and can supplement EHR data when it is available.

Improving Hospital Performance Rankings Using Discrete Patient Diagnoses for Risk Adjustment of Outcomes

OBJECTIVE

To assess the changes in patient outcome prediction and hospital performance ranking when incorporating diagnoses as risk adjusters rather than comorbidity indices.

DATA SOURCES

Healthcare Cost and Utilization Project State Inpatient Databases for New York State, 2005-2009.

STUDY DESIGN

Conducted tree-based classification for mortality and readmission by incorporating discrete patient diagnoses as predictors, comparing with traditional comorbidity indices such as those used for Centers for Medicare and Medicaid Services (CMS) outcome models.

PRINCIPAL FINDINGS

Diagnosis codes as predictors increased predictive accuracy 5.6 percent (95% CI: 4.5-6.9 percent) relative to CMS condition categories for heart failure 30-day mortality. Most other outcomes exhibited statistically significant accuracy gains and facility ranking shifts. Sensitivity analysis showed improvements even when predictors were limited to only the diagnoses included in CMS models.

CONCLUSIONS

Discretizing patient severity information beyond the levels of traditional comorbidity indices improves patient outcome predictions and substantially shifts facility rankings.

Regression Trees Identify Relevant Interactions: Can This Improve the Predictive Performance of Risk Adjustment?

Risk equalization formulas have been refined since their introduction about two decades ago. Because of the complexity and the abundance of possible interactions between the variables used, hardly any interactions are considered. A regression tree is used to systematically search for interactions, a methodologically new approach in risk equalization. Analyses are based on a data set of nearly 2.9 million individuals from a major German social health insurer. A two-step approach is applied: In the first step a regression tree is built on the basis of the learning data set. Terminal nodes characterized by more than one morbidity-group-split represent interaction effects of different morbidity groups. In the second step the 'traditional' weighted least squares regression equation is expanded by adding interaction terms for all interactions detected by the tree, and regression coefficients are recalculated. The resulting risk adjustment formula shows an improvement in the adjusted R² from 25.43% to 25.81% on the evaluation data set. Predictive ratios are calculated for subgroups affected by the interactions. The R² improvement detected is only marginal. According to the sample level performance measures used, not involving a considerable number of morbidity interactions forms no relevant loss in accuracy. Copyright © 2015 John Wiley & Sons, Ltd.

Real-time prediction of inpatient length of stay for discharge prioritization

OBJECTIVE

Hospitals are challenged to provide timely patient care while maintaining high resource utilization. This has prompted hospital initiatives to increase patient flow and minimize nonvalue added care time. Real-time demand capacity management (RTDC) is one such initiative whereby clinicians convene each morning to predict patients able to leave the same day and prioritize their remaining tasks for early discharge. Our objective is to automate and improve these discharge predictions by applying supervised machine learning methods to readily available health information.

MATERIALS AND METHODS

The authors use supervised machine learning methods to predict patients' likelihood of discharge by 2 p.m. and by midnight each day for an inpatient medical unit. Using data collected over 8000 patient stays and 20 000 patient days, the predictive performance of the model is compared to clinicians using sensitivity, specificity, Youden's Index (i.e., sensitivity + specificity - 1), and aggregate accuracy measures.

RESULTS

The model compared to clinician predictions demonstrated significantly higher sensitivity (P < .01), lower specificity (P < .01), and a comparable Youden Index (P > .10). Early discharges were less predictable than midnight discharges. The model was more accurate than clinicians in predicting the total number of daily discharges and capable of ranking patients closest to future discharge.

CONCLUSIONS

There is potential to use readily available health information to predict daily patient discharges with accuracies comparable to clinician predictions. This approach may be used to automate and support daily RTDC predictions aimed at improving patient flow.

Statin Use and the Risk of Kidney Disease With Long-Term Follow-Up (8.4-Year Study)

Few studies have examined long-term effects of statin therapy on kidney diseases. The objective of this study was to determine the association of statin use with incidence of acute and chronic kidney diseases after prolonged follow-up. In this retrospective cohort study, we analyzed data from the San Antonio area military health care system from October 2003 through March 2012. Statin users were propensity score matched to nonusers using 82 baseline characteristics including demographics, co-morbidities, medications, and health care utilization. Study outcomes were acute kidney injury, chronic kidney disease (CKD), and nephritis/nephrosis/renal sclerosis. Of the 43,438 subjects included, we propensity score matched 6,342 statin users with 6,342 nonusers. Statin users had greater odds of acute kidney injury (odds ratio [OR] 1.30, 95% confidence interval [CI] 1.14 to 1.48), CKD (OR 1.36, 95% CI 1.22 to 1.52), and nephritis/nephrosis/renal sclerosis (OR 1.35, 95% CI 1.05 to 1.73). In a subset of patients without co-morbidities, the association of statin use with CKD remained significant (OR 1.53, 95% CI 1.27 to 1.85). In a secondary analysis, adjusting for diseases/conditions that developed during follow-up weakened this association. In conclusion, statin use is associated with increased incidence of acute and chronic kidney disease. These findings are cautionary and suggest that long-term effects of statins in real-life patients may differ from shorter term effects in selected clinical trial populations.

Statins and New-Onset Diabetes Mellitus and Diabetic Complications: A Retrospective Cohort Study of US Healthy Adults

BACKGROUND

Statin use is associated with increased incidence of diabetes and possibly with increased body weight and reduced exercise capacity. Data on the long-term effects of these associations in healthy adults, however, are very limited. In addition, the relationship between these effects and diabetic complications has not been adequately studied.

OBJECTIVE

To examine the association between statin use and new-onset diabetes, diabetic complications, and overweight/obesity in a cohort of healthy adults.

RESEARCH DESIGN

This was a retrospective cohort study.

PARTICIPANTS

Subjects were Tricare beneficiaries who were evaluated between October 1, 2003 and March 1, 2012. Patients were divided into statin users and nonusers.

INTERVENTION

We excluded patients who, at baseline, had a preexisting disease indicative of cardiovascular diseases, any positive element of the Charlson comorbidity index (including diabetes mellitus), or life-limiting chronic diseases. Using 42 baseline characteristics, we generated a propensity score to match statin users and nonusers.

MAIN MEASURES

Outcomes assessed included new-onset diabetes, diabetic complications, and overweight/obesity.

KEY RESULTS

A total of 25,970 patients (3982 statin users and 21,988 nonusers) were identified as healthy adults at baseline. Of these, 3351 statins users and 3351 nonusers were propensity score-matched. Statin users had higher odds of new-onset diabetes (odds ratio [OR] 1.87; 95 % confidence interval [95 % CI] 1.67-2.01), diabetes with complications (OR 2.50; 95 % CI 1.88-3.32), and overweight/obesity (OR 1.14; 95 % CI 1.04-1.25). Secondary and sensitivity analyses demonstrated similar findings.

CONCLUSIONS

Diabetes, diabetic complications, and overweight/obesity were more commonly diagnosed among statin-users than similar nonusers in a healthy cohort of adults. This study demonstrates that short-term clinical trials might not fully describe the risk/benefit of long-term statin use for primary prevention.

Observing versus Predicting: Initial Patterns of Filling Predict Long-Term Adherence More Accurately Than High-Dimensional Modeling Techniques

OBJECTIVE

Despite the proliferation of databases with increasingly rich patient data, prediction of medication adherence remains poor. We proposed and evaluated approaches for improved adherence prediction.

DATA SOURCES

We identified Medicare beneficiaries who received prescription drug coverage through CVS Caremark and initiated a statin.

STUDY DESIGN

A total of 643 variables were identified at baseline from prior claims and linked Census data. In addition, we identified three postbaseline predictors, indicators of adherence to statins during each of the first 3 months of follow-up. We estimated 10 models predicting subsequent adherence, using logistic regression and boosted logistic regression, a nonparametric data-mining technique. Models were also estimated within strata defined by the index days supply.

RESULTS

In 77,703 statin initiators, prediction using baseline variables only was poor with maximum cross-validated C-statistics of 0.606 and 0.577 among patients with index supply ≤30 days and >30 days, respectively. Using only indicators of initial statin adherence improved prediction accuracy substantially among patients with shorter initial dispensings (C = 0.827/0.518), and, when combined with investigator-specified variables, prediction accuracy was further improved (C = 0.842/0.596).

CONCLUSIONS

Observed adherence immediately after initiation predicted future adherence for patients whose initial dispensings were relatively short.

Sex differences in the incidence of severe pain events following surgery: a review of 333,000 pain scores

OBJECTIVE/BACKGROUND

Prior work has not addressed sex differences in the incidence of severe postoperative pain episodes. The goal of this study was to examine sex differences in clinical postoperative pain scores across an array of surgical procedures using direct comparisons of numeric rating scale pain scores as well as using the incidence of severe pain events (SPEs).

DESIGN/SETTING

Retrospective cohort study of over 300,000 clinical pain score observations recorded from adult patients undergoing nonambulatory surgery at a tertiary care academic medical center over a 1-year period.

METHODS/PATIENTS

To test the hypothesis that the number of SPE on postoperative day (POD) 1 differed by sex after controlling for procedure, we calculated Cochran-Mantel-Haenszel statistics of sex by count of SPE, controlling for type of surgery.

ASSESSMENT TOOLS/OUTCOMES

Pain scores were collected from clinical nursing records where they were documented using the numeric rating scale.

RESULTS

In female patients, 10,989 (25.09%) of 43,806 POD 1 pain scores were considered SPE compared with 10,786 (22.45%) of 48,055 POD 1 pain scores in male patients. This produced an overall odds ratio of 1.16 (99% confidence interval 1.11-1.20) for females vs males to report an SPE for a pain score on POD 1. Estimates of the odds that a given pain observation represents an SPE for female vs male patients after controlling for type of surgery yielded an odds ratio of 1.14 (99% confidence interval, 1.10-1.19).

CONCLUSION

Female patients experience greater mean pain scores, as well as a higher incidence of SPE, on POD 1 for a variety of surgical procedures.

Classification tree analysis of race-specific subgroups at risk for a central venous catheter-related bloodstream infection

BACKGROUND

Studies of racial disparities in patient safety events often do not use race-specific risk adjustment and do not account for reciprocal covariate interactions. These limitations were addressed by using classification tree analysis separately for black patients and white patients to identify characteristics that segment patients who have increased risks for a venous catheter-related bloodstream infection.

METHODS

A retrospective, cross-sectional analysis of 5,236,045 discharges from 103 Florida acute hospitals in 2005-2009 was conducted. Hospitals were rank ordered on the basis of the black/white Patient Safety Indicator (PSI) 7 rate ratio as follows: Group 1 (white rate higher), Group 2, (equivalent rates), Group 3, (black rate higher), and Group 4, (black rate highest). Predictor variables included 26 comorbidities (Elixhauser Comorbidity Index) and demographic characteristics. Four separate classification tree analyses were completed for each race/hospital group.

RESULTS

Individual characteristics and groups of characteristics associated with increased PSI 7 risk differed for black and white patients. The average age for both races was different across the hospital groups (p < .01). Weight loss was the strongest single delineator and common to both races. The black subgroups with the highest PSI 7 risk were Medicare beneficiaries who were either < or = 25.5 years without hypertension or < or = 39.5 years without hypertension but with an emergency or trauma admission. The white subgroup with the highest PSI 7 risk consisted of patients < or = 45.5 years who had congestive heart failure but did not have either hypertension or weight loss.

DISCUSSION

Identifying subgroups of patients at risk for a rare safety event such as PSI 7 should aid effective clinical decisions and efficient use of resources and help to guide patient safety interventions.

A roadmap for designing a personalized search tool for individual healthcare providers

Each year, a large percentage of people change their physicians and other individual healthcare providers (IHPs). Many of these people have difficulty identifying a replacement they like. To help people find satisfactory IHPs who are likely to be good at managing their health issues and serve their needs well, in a previous paper we proposed a high-level framework for building a personalized search tool for IHPs. There are many issues regarding designing a personalized search tool for IHPs, of which only a small portion are mentioned in our previous paper. This paper surveys various such issues that are not covered in our previous paper. We include some preliminary thoughts on how to address these issues with the hope to stimulate future research work on the new topic of personalized search for IHPs.

Regression models for analyzing costs and their determinants in health care: an introductory review

OBJECTIVE

This article aims to describe the various approaches in multivariable modelling of healthcare costs data and to synthesize the respective criticisms as proposed in the literature.

METHODS

We present regression methods suitable for the analysis of healthcare costs and then apply them to an experimental setting in cardiovascular treatment (COSTAMI study) and an observational setting in diabetes hospital care.

RESULTS

We show how methods can produce different results depending on the degree of matching between the underlying assumptions of each method and the specific characteristics of the healthcare problem.

CONCLUSIONS

The matching of healthcare cost models to the analytic objectives and characteristics of the data available to a study requires caution. The study results and interpretation can be heavily dependent on the choice of model with a real risk of spurious results and conclusions.

Does nonresponse bias the results of retrospective surveys of end-of-life care?

OBJECTIVES

To evaluate the effect of nonresponse bias on reports of the quality of end-of-life care that older adults receive.

DESIGN

Nationwide retrospective survey of end-of-life care.

SETTING

Sixty-two Veterans Affairs Medical Centers.

PARTICIPANTS

Patients were eligible if they died in a participating facility. One family member per patient was selected from medical records and invited to participate.

MEASUREMENTS

The telephone survey included 14 items describing important aspects of the patient's care in the last month of life. Scores (0-100) reflect the percentage of items for which the family member reported that the patient received the best possible care, and a global item defined the proportion of families who said the patient received "excellent" care. To examine the effect of nonresponse bias, a model was created to predict the likelihood of response based on patient and family characteristics; then this model was used to apply weights that were equivalent to the inverse of the probability of response for that individual.

RESULTS

Interviews were completed with family members of 3,897 of 7,110 patients (55%). Once results were weighted to account for nonresponse bias, the change in mean individual scores was 2% of families reporting "excellent" care. Of the 62 facilities in the sample, the scores of only 19 facilities (31%) changed more than 1% in either direction, and only 10 (16%) changed more than 2%.

CONCLUSION

Although nonresponse bias is a theoretical concern, it does not appear to have a significant effect on the facility-level results of this retrospective family survey.