Propensity score matching for estimating a marginal hazard ratio.

Propensity score matching is commonly used to draw causal inference from observational survival data. However, its asymptotic properties have yet to be established, and variance estimation is still open to debate. We derive the statistical properties of the propensity score matching estimator of the marginal causal hazard ratio based on matching with replacement and a fixed number of matches. We also propose a double-resampling technique for variance estimation that takes into account the uncertainty due to propensity score estimation prior to matching.

Reweighting estimators to extend the external validity of clinical trials: methodological considerations.

Methods to extend the strong internal validity of randomized controlled trials to reliably estimate treatment effects in target populations are gaining attention. This paper enumerates steps recommended for undertaking such extended inference, discusses currently viable choices for each one, and provides recommendations. We demonstrate a complete extended inference from a clinical trial studying a pharmaceutical treatment for Alzheimer's disease (AD) to a realistic target population of European residents diagnosed with AD. This case study highlights approaches to overcoming practical difficulties and demonstrates limitations of reliably extending inference from a trial to a real-world population.

Practical recommendations on double score matching for estimating causal effects.

Unlike in randomized clinical trials (RCTs), confounding control is critical for estimating the causal effects from observational studies due to the lack of treatment randomization. Under the unconfoundedness assumption, matching methods are popular because they can be used to emulate an RCT that is hidden in the observational study. To ensure the key assumption hold, the effort is often made to collect a large number of possible confounders, rendering dimension reduction imperative in matching. Three matching schemes based on the propensity score (PSM), prognostic score (PGM), and double score (DSM, ie, the collection of the first two scores) have been proposed in the literature. However, a comprehensive comparison is lacking among the three matching schemes and has not made inroads into the best practices including variable selection, choice of caliper, and replacement. In this article, we explore the statistical and numerical properties of PSM, PGM, and DSM via extensive simulations. Our study supports that DSM performs favorably with, if not better than, the two single score matching in terms of bias and variance. In particular, DSM is doubly robust in the sense that the matching estimator is consistent requiring either the propensity score model or the prognostic score model is correctly specified. Variable selection on the propensity score model and matching with replacement is suggested for DSM, and we illustrate the recommendations with comprehensive simulation studies. An R package is available at https://github.com/Yunshu7/dsmatch.

A first step towards best practice recommendations for the design and statistical analyses of pragmatic clinical trials: A modified Delphi approach.

AIM: Pragmatic clinical trials (PCTs) are randomized trials implemented through routine clinical practice, where design parameters of traditional randomized controlled trials are modified to increase generalizability. However, this may introduce statistical challenges. We aimed to identify these challenges and discuss possible solutions leading to best practice recommendations for the design and analysis of PCTs. METHODS: A modified Delphi method was used to reach consensus among a panel of 11 experts in clinical trials and statistics. Statistical issues were identified in a focused literature review and aggregated with insights and possible solutions from experts collected through a series of survey iterations. Issues were ranked according to their importance. RESULTS: Twenty-seven articles were included and combined with experts' insight to generate a list of issues categorized into participants, recruiting sites, randomization, blinding and intervention, outcome (selection and measurement) and data analysis. Consensus was reached about the most important issues: risk of participants' attrition, heterogeneity of "usual care" across sites, absence of blinding, use of a subjective endpoint and data analysis aligned with the trial estimand. Potential issues should be anticipated and preferably be addressed in the trial protocol. The experts provided solutions regarding data collection and data analysis, which were considered of equal importance. DISCUSSION: A set of important statistical issues in PCTs was identified and approaches were suggested to anticipate and/or minimize these through data analysis. Any impact of choosing a pragmatic design feature should be gauged in the light of the trial estimand.

Diagnosis, consultation, treatment, and impact of migraine in the US: Results of the OVERCOME (US) study.

OBJECTIVE: The ObserVational survey of the Epidemiology, tReatment and Care of MigrainE (OVERCOME; United States) study is a multicohort, longitudinal web survey that assesses symptomatology, consulting, diagnosis, treatment, and impact of migraine in the United States. BACKGROUND: Regularly updating population-based views of migraine in the United States provides a method for assessing the quality of ongoing migraine care and identifying unmet needs. METHODS: The OVERCOME (US) 2018 migraine cohort involved: (I) creating a demographically representative sample of US adults using quota sampling (n = 97,478), (II) identifying people with active migraine in the past year via a validated migraine diagnostic questionnaire and/or self-reported medical diagnosis of migraine (n = 24,272), and (III) assessing consultation, diagnosis, and treatment of migraine (n = 21,143). The current manuscript evaluated whether those with low frequency episodic migraine (LFEM; 0-3 monthly headache days) differed from other categories on outcomes of interest. RESULTS: Among the migraine cohort (n = 21,143), 19,888 (94.1%) met our International Classification of Headache Disorders, 3rd edition-based case definition of migraine and 12,905 (61.0%) self-reported a medical diagnosis of migraine. Respondents' mean (SD) age was 42.2 (15.0) years; 15,697 (74.2%) were women. Having at least moderate disability was common (n = 8965; 42.4%) and around half (n = 10,783; 51.0%) had consulted a medical professional for migraine care in the past year. Only 4792 (22.7%) of respondents were currently using a triptan. Overall, 8539 (40.4%) were eligible for migraine preventive medication and 3555 (16.8%) were currently using migraine preventive medication. Those with LFEM differed from moderate and high frequency episodic migraine and chronic migraine on nearly all measures of consulting, diagnosis, and treatment. CONCLUSION: The OVERCOME (US) 2018 cohort revealed slow but steady progress in diagnosis and preventive treatment of migraine. However, despite significant impact among the population, many with migraine have unmet needs related to consulting for migraine, migraine diagnosis, and getting potentially beneficial migraine treatment. Moreover, it demonstrated the heterogeneity and varying unmet needs within episodic migraine.

A comparison of reweighting estimators of average treatment effects in real world populations.

Regulatory agencies typically evaluate the efficacy and safety of new interventions and grant commercial approval based on randomized controlled trials (RCTs). Other major healthcare stakeholders, such as insurance companies and health technology assessment agencies, while basing initial access and reimbursement decisions on RCT results, are also keenly interested in whether results observed in idealized trial settings will translate into comparable outcomes in real world settings-that is, into so-called "real world" effectiveness. Unfortunately, evidence of real world effectiveness for new interventions is not available at the time of initial approval. To bridge this gap, statistical methods are available to extend the estimated treatment effect observed in a RCT to a target population. The generalization is done by weighting the subjects who participated in a RCT so that the weighted trial population resembles a target population. We evaluate a variety of alternative estimation and weight construction procedures using both simulations and a real world data example using two clinical trials of an investigational intervention for Alzheimer's disease. Our results suggest an optimal approach to estimation depends on the characteristics of source and target populations, including degree of selection bias and treatment effect heterogeneity.

Addressing unmeasured confounding in comparative observational research.

PURPOSE: Observational pharmacoepidemiological studies can provide valuable information on the effectiveness or safety of interventions in the real world, but one major challenge is the existence of unmeasured confounder(s). While many analytical methods have been developed for dealing with this challenge, they appear under-utilized, perhaps due to the complexity and varied requirements for implementation. Thus, there is an unmet need to improve understanding the appropriate course of action to address unmeasured confounding under a variety of research scenarios. METHODS: We implemented a stepwise search strategy to find articles discussing the assessment of unmeasured confounding in electronic literature databases. Identified publications were reviewed and characterized by the applicable research settings and information requirements required for implementing each method. We further used this information to develop a best practice recommendation to help guide the selection of appropriate analytical methods for assessing the potential impact of unmeasured confounding. RESULTS: Over 100 papers were reviewed, and 15 methods were identified. We used a flowchart to illustrate the best practice recommendation which was driven by 2 critical components: (1) availability of information on the unmeasured confounders; and (2) goals of the unmeasured confounding assessment. Key factors for implementation of each method were summarized in a checklist to provide further assistance to researchers for implementing these methods. CONCLUSION: When assessing comparative effectiveness or safety in observational research, the impact of unmeasured confounding should not be ignored. Instead, we suggest quantitatively evaluating the impact of unmeasured confounding and provided a best practice recommendation for selecting appropriate analytical methods.

Evaluating different strategies for estimating treatment effects in observational studies.

Since the introduction of the propensity score (PS), methods for estimating treatment effects with observational data have received growing attention in the literature. Recent research has added substantially to the number of available statistical approaches for controlling confounding in such analyses. However, researchers need guidance to decide on the optimal analytic strategy for any given scenario. To address this gap, we conducted simulations evaluating both well-established methods (regression, PS weighting, stratification, and matching) and more recently proposed approaches (tree-based methods, local control, entropy balancing, genetic matching, prognostic scoring). The simulation scenarios included tree-based and smooth regression models as true data-generation mechanisms. We evaluated an extensive number of analysis strategies combining different treatment choices and outcome models. Key findings include 1) the lack of a single best strategy across all potential scenarios; 2) the importance of appropriately addressing interactions in the treatment choice model and/or outcome model; and 3) a tree-structured treatment choice model and a polynomial outcome model with second-order interactions performed well. One limitation to this initial assessment is the lack of heterogeneous simulation scenarios allowing treatment effects to vary by patient.

Propensity score matching and subclassification in observational studies with multi-level treatments.

In this article, we develop new methods for estimating average treatment effects in observational studies, in settings with more than two treatment levels, assuming unconfoundedness given pretreatment variables. We emphasize propensity score subclassification and matching methods which have been among the most popular methods in the binary treatment literature. Whereas the literature has suggested that these particular propensity-based methods do not naturally extend to the multi-level treatment case, we show, using the concept of weak unconfoundedness and the notion of the generalized propensity score, that adjusting for a scalar function of the pretreatment variables removes all biases associated with observed pretreatment variables. We apply the proposed methods to an analysis of the effect of treatments for fibromyalgia. We also carry out a simulation study to assess the finite sample performance of the methods relative to previously proposed methods.

Estimating optimal treatment regimes via subgroup identification in randomized control trials and observational studies.

With new treatments and novel technology available, personalized medicine has become an important piece in the new era of medical product development. Traditional statistics methods for personalized medicine and subgroup identification primarily focus on single treatment or two arm randomized control trials. Motivated by the recent development of outcome weighted learning framework, we propose an alternative algorithm to search treatment assignments which has a connection with subgroup identification problems. Our method focuses on applications from clinical trials to generate easy to interpret results. This framework is able to handle two or more than two treatments from both randomized control trials and observational studies. We implement our algorithm in C++ and connect it with R. Its performance is evaluated by simulations, and we apply our method to a dataset from a diabetes study. Copyright © 2016 John Wiley & Sons, Ltd.

"A Bayesian sensitivity analysis to evaluate the impact of unmeasured confounding with external data: a real world comparative effectiveness study in osteoporosis".

PURPOSE: Observational studies are frequently used to assess the effectiveness of medical interventions in routine clinical practice. However, the use of observational data for comparative effectiveness is challenged by selection bias and the potential of unmeasured confounding. This is especially problematic for analyses using a health care administrative database, in which key clinical measures are often not available. This paper provides an approach to conducting a sensitivity analyses to investigate the impact of unmeasured confounding in observational studies. METHODS: In a real world osteoporosis comparative effectiveness study, the bone mineral density (BMD) score, an important predictor of fracture risk and a factor in the selection of osteoporosis treatments, is unavailable in the data base and lack of baseline BMD could potentially lead to significant selection bias. We implemented Bayesian twin-regression models, which simultaneously model both the observed outcome and the unobserved unmeasured confounder, using information from external sources. A sensitivity analysis was also conducted to assess the robustness of our conclusions to changes in such external data. RESULTS: The use of Bayesian modeling in this study suggests that the lack of baseline BMD did have a strong impact on the analysis, reversing the direction of the estimated effect (odds ratio of fracture incidence at 24 months: 0.40 vs. 1.36, with/without adjusting for unmeasured baseline BMD). CONCLUSIONS: The Bayesian twin-regression models provide a flexible sensitivity analysis tool to quantitatively assess the impact of unmeasured confounding in observational studies. Copyright © 2016 John Wiley & Sons, Ltd.

The importance of unresolved fatigue in depression: costs and comorbidities.

OBJECTIVE: To assess the cost outcomes of patients with a history of depression and clinically significant fatigue. METHODS: Adults with ≥ 2 claims with depression diagnosis codes identified from the HealthCore Integrated Research Database were invited to participate in this study linking survey data with retrospective claims data (12-mo presurvey and postsurvey periods). Patient surveys included measures for depression (Quick Inventory of Depressive Symptomatology), fatigue (Fatigue Associated with Depression Questionnaire), anxiety (7-item Generalized Anxiety Disorder scale), sleep difficulty (Athens Insomnia Scale), and pain (Brief Pain Inventory). After adjusting for demographic and clinical characteristics using propensity scores, postsurvey costs were compared between patients with and without fatigue using nonparametric bootstrapping methods. RESULTS: Of the 1982 patients who had completed the survey and had complete claims data, 653 patients had significant levels of fatigue. Patients with fatigue reported significantly higher scores, indicating greater severity, on measures of depression, pain, sleep difficulty, and anxiety (all p < 0.05). These patients also had higher levels of overall medication use and were more likely to have lower measures of socioeconomic status than patients without significant levels of fatigue (all p < 0.05). Mean annual total costs were greater for patients with fatigue than those without fatigue ($14,462 vs $9971, respectively, p < 0.001). These cost differences remained statistically significant after adjusting for clinical and demographic differences. CONCLUSIONS: Clinically significant fatigue appears to add to the economic burden of depression. This reinforces the need for aggressive treatment of all symptoms and further examination of the variability of this relationship as patients approach remission.

Baseline characteristics and hospitalizations in patients with schizophrenia receiving olanzapine long-acting injection: an interim analysis from a non-interventional, prospective observational safety study.

BACKGROUND: Depot antipsychotics are a treatment option for medication nonadherence in patients with schizophrenia. Nonadherence can lead to increased relapse and hospitalization rates. This article reports hospitalization data before and after initiation of olanzapine long-acting injection (LAI), a depot antipsychotic. METHODS: Data were assessed from an ongoing, multinational, prospective, observational post-authorisation safety study being conducted to evaluate post-injection delirium/sedation syndrome (PDSS), an adverse reaction that can occur following injection of olanzapine LAI. Eligible patients were aged ≥18 years, diagnosed with schizophrenia, were prescribed olanzapine LAI, and lived outside the United States. Psychiatric hospitalization and medication data were collected retrospectively for the 6-month period before study entry and prospectively throughout the study. Paired t-tests and McNemar's tests were used to assess changes in hospitalization incidence and duration. Stepwise Cox proportional hazards models assessed factors associated with hospitalizations. Analyses were based on data from the first 3 years of the continuously enrolling study (N = 668). RESULTS: The average duration of olanzapine LAI exposure for all patients was 0.768 years. Of the 529 patients who received at least 1 injection of olanzapine LAI and were not hospitalized at study entry, 8.1% had at least 1 subsequent psychiatric hospitalization with a mean duration of 2.0 days. Of the 288 patients who had a >6-month follow-up, 8.3% had at least 1 post-baseline psychiatric hospitalization with a mean duration of 2.3 days. The incidence of hospitalizations in the 6-month period after treatment was significantly lower than that in the 6-month period prior to treatment (8.3 vs 32.6%, respectively; P < 0.001). Furthermore, mean hospitalization duration decreased from 11.5 days in the 6-month period before treatment to 2.3 days in the 6-month period after treatment (P < 0.001). Psychiatric hospitalization in the prior 12 months (P < 0.0001) and recreational drug use within 24 h of baseline visit (P = 0.015) were identified as potential predictors of time to first psychiatric hospitalization after beginning to take olanzapine LAI. At the time of interim analysis, 5 PDSS events had occurred, which was too few for a full analysis of those events. CONCLUSIONS: Results indicate a significant reduction in the incidence and days of hospitalization from the 6-month period before to the 6-month period after olanzapine LAI initiation, which suggests reduced relapse and hospitalization during treatment. Results should be interpreted with caution due to the observational nature of the study and use of retrospective baseline data.

Adaptive truncated weighting for improving marginal structural model estimation of treatment effects informally censored by subsequent therapy.

Randomized clinical trials are designed to estimate the direct effect of a treatment by randomly assigning patients to receive either treatment or control. However, in some trials, patients who discontinued their initial randomized treatment are allowed to switch to another treatment. Therefore, the direct treatment effect of interest may be confounded by subsequent treatment. Moreover, the decision on whether to initiate a second-line treatment is typically made based on time-dependent factors that may be affected by prior treatment history. Due to these time-dependent confounders, traditional time-dependent Cox models may produce biased estimators of the direct treatment effect. Marginal structural models (MSMs) have been applied to estimate causal treatment effects even in the presence of time-dependent confounders. However, the occurrence of extremely large weights can inflate the variance of the MSM estimators. In this article, we proposed a new method for estimating weights in MSMs by adaptively truncating the longitudinal inverse probabilities. This method provides balance in the bias variance trade-off when large weights are inevitable, without the ad hoc removal of selected observations. We conducted simulation studies to explore the performance of different methods by comparing bias, standard deviation, confidence interval coverage rates, and mean square error under various scenarios. We also applied these methods to a randomized, open-label, phase III study of patients with nonsquamous non-small cell lung cancer.

Medical costs of Alzheimer's disease misdiagnosis among US Medicare beneficiaries.

INTRODUCTION: Recent developments in diagnostic technology can support earlier, more accurate diagnosis of non-Alzheimer's disease (AD) dementias. METHODS: To evaluate potential economic benefits of early rule-out of AD, annual medical resource use and costs for Medicare beneficiaries potentially misdiagnosed with AD prior to their diagnosis of vascular dementia (VD) or Parkinson's disease (PD) were compared with that of similar patients never diagnosed with AD. RESULTS: Patients with prior AD diagnosis used substantially more medical services every year until their VD/PD diagnosis, resulting in incremental annual medical costs of approximately $9,500-$14,000. However, following their corrected diagnosis, medical costs converged with those of patients never diagnosed with AD. DISCUSSION: The observed correlation between timing of correct diagnosis and subsequent reversal in excess costs is strongly suggestive of the role of misdiagnosis of AD - rather than AD comorbidity - in this patient population. Our findings suggest potential benefits from earlier, accurate diagnosis.

Bayesian modeling of cost-effectiveness studies with unmeasured confounding: a simulation study.

Unmeasured confounding is a common problem in observational studies. Failing to account for unmeasured confounding can result in biased point estimators and poor performance of hypothesis tests and interval estimators. We provide examples of the impacts of unmeasured confounding on cost-effectiveness analyses using observational data along with a Bayesian approach to correct estimation. Assuming validation data are available, we propose a Bayesian approach to correct cost-effectiveness studies for unmeasured confounding. We consider the cases where both cost and effectiveness are assumed to have a normal distribution and when costs are gamma distributed and effectiveness is normally distributed. Simulation studies were conducted to determine the impact of ignoring the unmeasured confounder and to determine the size of the validation data required to obtain valid inferences.

Direct medical costs and medication compliance among fibromyalgia patients: duloxetine initiators vs. pregabalin initiators.

OBJECTIVES: To assess and compare direct medical costs and medication compliance between patients with fibromyalgia who initiated duloxetine and patients with fibromyalgia who initiated pregabalin in 2008. METHODS: A retrospective cohort study design was used based on a large US national commercial claims database (2006 to 2009). Patients with fibromyalgia aged 18 to 64 who initiated duloxetine or pregabalin in 2008 and who had continuous health insurance 1 year preceding and 1 year following the initiation were selected into duloxetine cohort or pregabalin cohort based on their initiated agent. Medication compliance was measured by total supply days, medication possession ratio (MPR), and proportion of patients with MPR ≥ 0.8. Direct medical costs were measured by annual costs per patient and compared between the cohorts in the year following the initiation. Propensity score stratification and bootstrapping methods were used to adjust for distribution bias, as well as cross-cohort differences in demographic, clinical and economic characteristics, and medication history prior to the initiation. RESULTS: Both the duloxetine (n = 3,033) and pregabalin (n = 4,838) cohorts had a mean initiation age around 49 years, 89% were women. During the postindex year, compared to the pregabalin cohort, the duloxetine cohort had higher totally annual supply days (273.5 vs. 176.6, P < 0.05), higher MPR (0.7 vs. 0.5, P < 0.05), and more patients with MPR ≥ 0.8 (45.1% vs. 29.4%, P < 0.05). Further, relative to pregabalin cohort, duloxetine cohort had lower inpatient costs ($2,994.9 vs. $4,949.6, P < 0.05), lower outpatient costs ($8,259.6 vs. $10,312.2, P < 0.05), similar medication costs ($5,214.6 vs. $5,290.8, P > 0.05), and lower total medical costs ($16,469.1 vs. $20,552.6, P < 0.05) in the postinitiation year. CONCLUSIONS: In a real-world setting, patients with fibromyalgia who initiated duloxetine in 2008 had better medication compliance and consumed less inpatient, outpatient, and total medical costs than those who initiated pregabalin.

Evaluating the impact of unmeasured confounding with internal validation data: an example cost evaluation in type 2 diabetes.

The quantitative assessment of the potential influence of unmeasured confounders in the analysis of observational data is rare, despite reliance on the "no unmeasured confounders" assumption. In a recent comparison of costs of care between two treatments for type 2 diabetes using a health care claims database, propensity score matching was implemented to adjust for selection bias though it was noted that information on baseline glycemic control was not available for the propensity model. Using data from a linked laboratory file, data on this potential "unmeasured confounder" were obtained for a small subset of the original sample. By using this information, we demonstrate how Bayesian modeling, propensity score calibration, and multiple imputation can utilize this additional information to perform sensitivity analyses to quantitatively assess the potential impact of unmeasured confounding. Bayesian regression models were developed to utilize the internal validation data as informative prior distributions for all parameters, retaining information on the correlation between the confounder and other covariates. While assumptions supporting the use of propensity score calibration were not met in this sample, the use of Bayesian modeling and multiple imputation provided consistent results, suggesting that the lack of data on the unmeasured confounder did not have a strong impact on the original analysis, due to the lack of strong correlation between the confounder and the cost outcome variable. Bayesian modeling with informative priors and multiple imputation may be useful tools for unmeasured confounding sensitivity analysis in these situations. Further research to understand the operating characteristics of these methods in a variety of situations, however, remains.

Longitudinal observation of treatment patterns and outcomes for patients with fibromyalgia: 12-month findings from the reflections study.

OBJECTIVE: To describe 12-month treatment patterns and outcomes for patients starting a new medication for fibromyalgia in routine clinical practice. DESIGN AND OUTCOME MEASURES: Data from 1,700 patients were collected at baseline and 1, 3, 6, and 12 months. Repeated measures and Poisson regression models controlling for demographic, clinical, and baseline outcomes were used to assess changes in health outcomes (Brief Pain Inventory severity and interference, Sheehan Disability Scale, Fibromyalgia Impact Questionnaire), satisfaction, and economic factors for patients who initiated on pregabalin (214, 12.6%), duloxetine (264, 15.5%), milnacipran (134, 7.9%), or tricyclic antidepressants (66, 3.9%). Sensitivity analyses were run using propensity-matched cohorts. RESULTS: Patients started on 145 unique drugs for fibromyalgia, and over 75% of patients took two or more medications concurrently for fibromyalgia at each time point assessed. Overall, patients showed improvement on the four health outcomes, with few differences across medication cohorts. At baseline, patients reported annual averages of 20.3 visits for outpatient care, 27.7 missed days of work, and 32.6 days of care by an unpaid caregiver. The duloxetine and milnacipran (vs pregabalin or tricyclic antidepressant) cohorts had fewer outpatient visits during the 12-month study. Patients reported satisfaction with overall treatment and their fibromyalgia medication (46.0% and 42.8%, respectively). CONCLUSIONS: In this real-world setting, patients with fibromyalgia reported modest improvements, high resource, and medication use, and were satisfied with the care they received. Cohort differences were difficult to discern because of the high rates of drug discontinuation and concomitant medication use over the 12-month study period.

Predicting Response to Bolus Insulin Therapy in Patients With Type 2 Diabetes.

BACKGROUND: The aim of this study was to develop a predictive model to classify people with type 2 diabetes (T2D) into expected levels of success upon bolus insulin initiation. METHODS: Machine learning methods were applied to a large nationally representative insurance claims database from the United States (dNHI database; data from 2007 to 2017). We trained boosted decision tree ensembles (XGBoost) to assign people into Class 0 (never meeting HbA1c goal), Class 1 (meeting but not maintaining HbA1c goal), or Class 2 (meeting and maintaining HbA1c goal) based on the demographic and clinical data available prior to initiating bolus insulin. The primary objective of the study was to develop a model capable of determining at an individual level, whether people with T2D are likely to achieve and maintain HbA1c goals. HbA1c goal was defined at <8.0% or reduction of baseline HbA1c by >1.0%. RESULTS: Of 15 331 people with T2D (mean age, 53.0 years; SD, 8.7), 7800 (50.9%) people met HbA1c goal but failed to maintain that goal (Class 1), 4510 (29.4%) never attained this goal (Class 0), and 3021 (19.7%) people met and maintained this goal (Class 2). Overall, the model's receiver operating characteristic (ROC) was 0.79 with greater performance on predicting those in Class 2 (ROC = 0.92) than those in Classes 0 and 1 (ROC = 0.71 and 0.62, respectively). The model achieved high area under the precision-recall curves for the individual classes (Class 0, 0.46; Class 1, 0.58; Class 2, 0.71). CONCLUSIONS: Predictive modeling using routine health care data reasonably accurately classified patients initiating bolus insulin who would achieve and maintain HbA1c goals, but less so for differentiation between patients who never met and who did not maintain goals. Prior HbA1c was a major contributing parameter for the predictions.