Matching by Propensity Score in Cohort Studies with Three Treatment Groups

From the aDivision of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, MA
Epidemiology (Cambridge, Mass.) (Impact Factor: 6.2). 03/2013; 24(3). DOI: 10.1097/EDE.0b013e318289dedf
Source: PubMed


Nonrandomized pharmacoepidemiology generally compares one medication with another. For many conditions, clinicians can benefit from comparing the safety and effectiveness of three or more appropriate treatment options. We sought to compare three treatment groups simultaneously by creating 1:1:1 propensity score-matched cohorts.

We developed a technique that estimates generalized propensity scores and then creates 1:1:1 matched sets. We compared this methodology with two existing approaches-construction of matched cohorts through a common-referent group and a pairwise match for each possible contrast. In a simulation, we varied unmeasured confounding, presence of treatment effect heterogeneity, and the prevalence of treatments and compared each method's bias, variance, and mean squared error (MSE) of the treatment effect. We applied these techniques to a cohort of rheumatoid arthritis patients treated with nonselective nonsteroidal anti-inflammatory drugs, COX-2 selective inhibitors, or opioids.

We performed 1000 simulation runs. In the base case, we observed an average bias of 0.4% (MSE × 100 = 0.2) in the three-way matching approach and an average bias of 0.3% (MSE × 100 = 0.2) with the pairwise technique. The techniques showed differing bias and MSE with increasing treatment effect heterogeneity and decreasing propensity score overlap. With highly unequal exposure prevalences, strong heterogeneity, and low overlap, we observed a bias of 6.5% (MSE × 100 = 10.8) in the three-way approach and 12.5% (MSE × 100 = 12.3) in the pairwise approach. The empirical study displayed better covariate balance using the pairwise approach. Point estimates were substantially similar.

Our matching approach offers an effective way to study the safety and effectiveness of three treatment options. We recommend its use over the pairwise or common-referent approaches.

26 Reads
  • [Show abstract] [Hide abstract]
    ABSTRACT: To determine practice patterns and perioperative outcomes for open and minimally invasive partial nephrectomy (PN) in the United States since the introduction of a robot-assisted modifier in the Nationwide Inpatient Sample (NIS). Relying on the NIS between October 2008-December 2010, all patients with non-metastatic disease undergoing open PN (OPN), laparoscopic PN (LPN) and robotic PN (RPN) were identified. Utilization rates were assessed according to year, patient and hospital characteristics. Peri-operative outcomes between OPN vs. RPN, and OPN vs. LPN were evaluated using binary logistic regression models adjusted for patient and hospital co-variates. In a weighted sample of 38064 PNs, 66.9%, 23.9% and 9.2% were OPN, RPN and LPN, respectively. In 2010, the relative annual increase in the utilization of OPN, RPN and LPN was 7.9%, 45.4% and 6.1% respectively. Compared to OPN, patients undergoing minimally invasive PN were less likely to receive a blood transfusion (RPN odds ratio [OR]:0.56, p<0.001; LPN OR:0.68, p=0.016), postoperative complication (RPN OR:0.63, p<0.001; LPN OR:0.78, p<0.009), or prolonged length of stay (RPN OR:0.27, p<0.001; LPN OR:0.41, p<0.001). Only RPN patients were less likely to experience an intraoperative complication (RPN OR:0.69, p=0.014; LPN OR:0.67, p=0.069). Excess hospital charges were higher after RPN (OR:1.35, p<0.001). The dissemination of robotic surgery for PN in the U.S. has been rapid and safe. Compared to OPN, RPN had lower odds than LPN for most studied outcomes except hospital charges. RPN has now supplanted LPN as the most common minimally invasive approach for PN.
    The Journal of urology 10/2013; 191(4). DOI:10.1016/j.juro.2013.10.099 · 4.47 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Objectives: The study objective was to evaluate the effects of sitagliptin in patients with type 2 diabetes (T2D) andheart failure (HF). Background: There is uncertainty in the literature about whether dipeptidyl peptidase (DPP)-4 inhibitors cause harmin patients with HF and T2D. Methods: We analyzed data from a national commercially insured U.S. claims database. Patients with incident HF were identified from individuals with T2D initially treated with metformin or sulfonylurea and followed over time. Subjects subsequently using sitagliptin were compared with those not using sitagliptin in the 90 days before our primary outcome of all-cause hospital admission or death using a nested case-control analysis after adjustment for demographics and clinical and laboratory data. HF-specific hospital admission or death also was assessed. Results: A total of 7,620 patients with diabetes and incident HF met our inclusion criteria. Mean (SD) age was 54 years (9), and 58% (3,180) were male. Overall, 887 patients (12%) were exposed to sitagliptin therapy (521 patient years ofexposure) after incident HF. Our primary composite endpoint occurred in 4,137 patients (54%). After adjustment, sitagliptin users were not at an increased risk for the primary endpoint (7.1% vs. 9.2%, adjusted odds ratio [aOR]: 0.84, 95% confidence interval [CI]: 0.69 to 1.03) or each component (hospital admission 7.5% vs. 9.2%, aOR: 0.93, 95% CI: 0.76 to 1.14; death 6.9% vs. 9.3%, aOR: 1.16, 95% CI: 0.68 to 1.97). However, sitagliptin use was associated with an increased risk of HF hospitalizations (12.5% vs. 9.0%, aOR: 1.84, 95% CI: 1.16 to 2.92). Conclusions: Sitagliptin use was not associated with an increased risk of all-cause hospitalizations or death, but wasassociated with an increased risk of HF-related hospitalizations among patients with T2D with pre-existing HF.
    06/2014; 2(6). DOI:10.1016/j.jchf.2014.04.005
  • [Show abstract] [Hide abstract]
    ABSTRACT: Research to date within the field of well-being program evaluation has considered the study population to have either been given a treatment or not, and that matching will yield an unbiased, efficient estimate of the treatment causal effect. As well-being intervention programs become more sophisticated and diverse in their offerings, so too must the methods for assessing program effect. The objective of this research was to extend the traditional binary cohort assignment in quasi-experimental program evaluation in order to quantify the differential effects of a multi-tiered well-being improvement program administered over a 3 year period in a large employer. Data collected over this 3 year period included well-being assessments and medical claims from 17,669 employees and spouses. These individuals were assigned different cohorts based on intervention program intensity and matched utilizing coarsened exact matching. The matching process was able to remove 85 %, on average, of detectable bias across all comparison cohorts. A weighted generalized linear model, using the coarsened exact matching derived weights, was estimated to quantify the net (difference-in-difference) causal effect of the well-being intervention program. The results showed an increase of overall well-being on average in the High Intensity cohort of 1.48 and 1.32 points in the Mild Intensity cohort. The non-intervened cohort only evidenced a 0.57 point increase in overall well-being. The methodology reported here provides an expanded and robust approach to matching on different cohorts for the purpose of program evaluation.
    Health Services and Outcomes Research Methodology 06/2015; 15(2). DOI:10.1007/s10742-014-0136-7