Comparison of Nearest Neighbor and Caliper Algorithms in Outcome Propensity Score Matching to Study the Relationship between Type 2 Diabetes and Coronary Artery Disease

Abstract

Introduction: Propensity score matching (PSM) is a method to reduce the impact of essential and confounders. When the number of confounders is high, there may be a problem of matching, in which, finding matched pairs for the case group is difficult, or impossible. The propensity score (PS) minimizes the effect of the confounders, and it is reduced to one dimension. There are various algorithms in the field of PSM. This study aimed to compared the nearest neighbor and caliper algorithms.
 Methods: Data obtained in this study were from patients undergoing angiography at Ghaem Hospital in Mashhad, between 2011-12. The study was a retrospective case-control using PSM. In total, 604 patients were included in the case and control groups. A logistic regression model was used to calculate the propensity score and adjust the variables, such as age, gender, Body Mass Index (BMI), systolic blood pressure, smoking status, and triglyceride. Then, the Odds Ratios (ORs) with 95% Confidence Intervals (CIs) for the raw data and two matching algorithms were determined to examine the relationship between type 2 diabetes and coronary artery disease (CAD).
 Results: Propensity score in the nearest neighbor and caliper algorithms matched the total number of 604 samples, 200 and 178 pairs, respectively. All variables were significantly different between the two groups before matching (P<0.05). The gender was significantly different between the two groups after matching using the nearest neighbor algorithm (P=0.002). No variables created a significant difference between the two groups after matching with the caliper algorithm.
 Conclusion: Bias reduction in the caliper algorithm was greater than for the nearest neighbor algorithm for all variables except the triglyceride variable.

Optimized Weighted Nearest Neighbours Matching Algorithm for Control Group Selection

“…Although different methods may produce control groups of different quality [15] (depending on the nature of the population used for selection), studies published in the literature most often only use the greedy PSM matching. Despite the popularity of the widely applied greedy k-nn-based PSM method, it has also got many criticisms [16][17][18][19][20][21][22].…”