Iterative Causal Forest: A Novel Algorithm for Subgroup Identification.

Precisely and efficiently identifying subgroups with heterogeneous treatment effects (HTEs) in real-world evidence studies remains a challenge. Based on the causal forest (CF) method, we developed an iterative CF (iCF) algorithm to identify HTEs in subgroups defined by important variables. Our method iteratively grows different depths of the CF with important effect modifiers, performs plurality votes to obtain decision trees (subgroup decisions) for a family of CFs with different depths, then finds the cross-validated subgroup decision that best predicts the treatment effect as a final subgroup decision. We simulated 12 different scenarios and showed that the iCF outperformed other machine learning methods for interaction/subgroup identification in the majority of scenarios assessed. Using a 20% random sample of fee-for-service Medicare beneficiaries initiating sodium-glucose cotransporter-2 inhibitors (SGLT2i) or glucagon-like peptide-1 receptor agonists (GLP1RA), we implemented the iCF to identify subgroups with HTEs for hospitalized heart failure. Consistent with previous studies suggesting patients with heart failure benefit more from SGLT2i, iCF successfully identified such a subpopulation with HTEs and additive interactions. The iCF is a promising method for identifying subgroups with HTEs in real-world data where the potential for unmeasured confounding can be limited by study design.

The Breast and Cervical Cancer Early Detection Program, Medicaid, and breast cancer outcomes among Ohio's underserved women.

BACKGROUND: As an organized screening program, the national Breast and Cervical Cancer Early Detection Program (BCCEDP) was launched in the early 1990s to improve breast cancer outcomes among underserved women. To analyze the impact of the BCCEDP on breast cancer outcomes in Ohio, this study compared cancer stages and mortality across BCCEDP participants, Medicaid beneficiaries, and "all others." METHODS: This study linked data across the Ohio Cancer Incidence Surveillance System, Medicaid, the BCCEDP database, death certificates, and the US Census and identified 26,426 women aged 40 to 64 years who had been diagnosed with incident invasive breast cancer during the years 2002-2008 (deaths through 2010). The study groups were as follows: BCCEDP participants (1-time or repeat users), Medicaid beneficiaries (women enrolled in Medicaid before their cancer diagnosis [Medicaid/prediagnosis] or around the time of their cancer diagnosis [Medicaid/peridiagnosis]), and all others (women identified as neither BCCEDP participants nor Medicaid beneficiaries). The outcomes included advanced-stage cancer at diagnosis and mortality. A multivariable logistic and survival analysis was conducted to examine the independent association between the BCCEDP and Medicaid status and the outcomes. RESULTS: The percentage of women presenting with advanced-stage disease was highest among women in the Medicaid/peridiagnosis group (63.4%) and lowest among BCCEDP repeat users (38.6%). With adjustments for potential confounders and even in comparison with Medicaid/prediagnosis beneficiaries, those in the Medicaid/peridiagnosis group were twice as likely to be diagnosed with advanced-stage disease (adjusted odds ratio, 2.20; 95% confidence interval, 1.83-2.66). CONCLUSIONS: Medicaid/peridiagnosis women are at particularly high risk to be diagnosed with advanced-stage disease. Efforts to reduce breast cancer disparities must target this group of women before they present to Medicaid. Cancer 2017;123:3097-106. © 2017 American Cancer Society.

Health Care Utilization and Costs Associated With Empagliflozin in Older Adults With Type 2 Diabetes.

OBJECTIVE: We compared health care resource utilization (HCRU) and costs for inpatient and outpatient services and dispensed medications in older adults with type 2 diabetes initiating empagliflozin versus dipeptidyl peptidase 4 inhibitors (DPP-4is). RESEARCH DESIGN AND METHODS: The study population included U.S. Medicare fee-for-service beneficiaries with diabetes (age ≥65 years) initiating empagliflozin or DPP-4is (August 2014 to September 2018). We estimated rate ratios (RRs) for HCRU outcomes using zero-inflated negative binomial regression and per-member per-year (PMPY) cost differences using generalized linear model with gamma distributions, overall and stratified by baseline cardiovascular disease (CVD), after adjusting for 143 baseline covariates using 1:1 propensity score matching. RESULTS: We identified 23,335 matched pairs (mean age 72 years, 51% with baseline CVD). HCRU rates were lower in empagliflozin versus DPP-4i initiators (number of inpatient days, RR 0.89 [95% CI 0.82, 0.97]; number of emergency department [ED] visits, 0.86 [0.82, 0.91]; number of hospitalizations, 0.86 [0.79, 0.93]; number of office visits, 0.96 [0.95, 0.98]). Inpatient cost (-$713 PMPY [95% CI -847, -579), outpatient cost (-$198 PMPY[-272, -124]), and total cost of care (-$1,109 PMPY [-1,478, -739]) were lower for empagliflozin versus DPP-4is, although diabetes medication cost was higher in empagliflozin initiators ($454 PMPY [95% CI 284, 567]). In the CVD subgroup, total cost was lower for empagliflozin initiators (-$2,005 PMPY [-2,451, -1,337]), while the difference was attenuated in the non-CVD subgroup (-$296 PMPY[-740, 148]). CONCLUSIONS: Among older adults with diabetes, empagliflozin was associated with a lower number of inpatient days, hospitalizations, ED visits, and office visits and with lower costs of care compared with DPP-4is, especially in those with history of CVD.

A comparison of the effects of epidural analgesia versus traditional pain management on outcomes after gastric cancer resection: a population-based study.

BACKGROUND AND OBJECTIVES: Epidural analgesia may increase survival after cancer surgery by reducing recurrence. This population-based study compared survival and treated recurrence after gastric cancer resection between patients receiving epidurals and those who did not. METHODS: We used the linked federal Surveillance, Epidemiology, and End Results Program/Medicare database to identify patients aged 66 years or older with nonmetastatic gastric carcinoma diagnosed 1996 to 2005 who underwent resection. Exclusions included diagnosis at autopsy, no Medicare Part B, familial cancer syndrome, emergency surgery, and laparoscopic procedures. Epidurals were identified by Current Procedural Terminology codes. Treated recurrence was defined as chemotherapy greater than or equal to 16 months and/or radiation greater than or equal to 12 months after surgery. Recurrence was compared by conditional logistic regression. Survival was compared via marginal Cox proportional hazards regression model. RESULTS: We identified 2745 patients, 766 of whom had epidural codes. Patients receiving epidurals were more likely to have regional disease, be white, and live in areas with relatively high socioeconomic status. Overall treated recurrence was 25.6% (27.5% epidural and 24.9% nonepidural). In the adjusted logistic regression, there was no difference in recurrence (odds ratio, 1.40; 95% confidence interval [CI], 0.96-2.05). Median survival did not differ: 28.1 months (95% CI, 24.8-32.3) in the epidural versus 27.4 months (95% CI, 24.8-30.0) in the nonepidural groups. The marginal Cox models showed no association between epidural use and mortality (adjusted hazard ratio, 0.93; 95% CI, 0.84-1.03). CONCLUSIONS: There was no difference between groups regarding treated recurrence or survival. Whether this is true or simply a result of insufficient power is unclear. Prospective studies are needed to provide stronger evidence.