Propensity+Score+Matching+in+SPSS: …

1 propensity score matching in SPSS:How to turn an Audit into a RCTO utline What is propensity score matching ? propensity score matching in SPSS Example: Comparing patients with both Gout & diabetes to those with diabetes only Dealing with missing dataMario D Hair Independent Statistics Consultant1 Mario D Hair Independent Statistics Consultant What is propensity score matching ?Developed by Rosenbaum & Rubin (1983). Two aspects1. Generate the propensity score2. Apply it to balance the hits using propensity score matching by year.

2 Slide provided by Beng So, ST6 Queen Elizabeth Hospital, GlasgowMario D Hair Independent Statistics Consultant2 What is propensity score matching ?1. Generate the propensity scoreThe propensity score is the probability (from 0 to 1) of a case being in a particular group based on a given set of covariates. Generally calculated using logistic regression with group (Treatment /Control) as dependent , covariates as independent variables. Caveats & Limitations Can only be two groups. If more groups need to analyse them pairwise.

3 The propensity score is only as good as the predictors used to generate it. propensity score not generated for any case with any missing data. Not interested in any aspect of the logistic model other than the probabilities. The propensity score is a balancing score : The differences between groups on the covariates condensed down into a single score so if two groups balanced on the propensity score then balanced on all the D Hair Independent Statistics Consultant3 Mario D Hair Independent Statistics Consultant4 Slide provided by Beng So, ST6 Queen Elizabeth Hospital, GlasgowWhat is propensity score matching ?

4 2. Apply propensity score to balance the data. Four main score matching : Match one or more control cases with a propensity score that is (nearly) equal to the propensity score for each treatment case Stratification:Divide sample into strata based on rank- ordered propensity scores. Comparisons between groups are then performed within each adjustment:Include propensity scores as a covariate in a regression model used to estimate the treatment effect. Weighting: Inverse probability of treatment weighting (IPTW) weights cases by the inverse of propensity score .

5 Similar to use of survey sampling weights used to ensure samples are representative of specific populations. Often used in survival (2011) reports that propensity score matching is better than stratification or regression adjustment and is at least as good as IPTW. It is increasingly the most widely used D Hair Independent Statistics Consultant5 propensity score matching in SPSSA vailable in SPSSV22 but Prior to that only as PS matching an extension command that requires both r and the r plug- in.

6 Developed by Felix Thoemmes at Cornell University. PS matching : contains Latest version of the software, psmatching June 2015. (this talk uses ) Installation instructions (in a file called ) Thoemmes 2012 paper describing the software (called arxiv ).Comparison of PS matching & SPSS propensity score matchingMario D Hair Independent Statistics Consultant6 PS matching SPSS propensity score matching Loading Can be tricky. Requires R plug- in & R but available for V18 onwards Pre- loaded in V22 Generate propensity score SPSS logistic regression GAM logit?

7 ? SPSS logistic regression score matching Uses R packages: MatchIt, Rltools , cem Uses Python essentials FUZZY extension command Speed Can be slow for large files Also slow but speed can be increased by sacrificing precision Precision Very good Can be poor unless match tolerance set very low Diagnostics Very good Poor Missing data Cannot handle any missing data, covariate or not. No problem but missing data in the covariates will result in omission of cases Mario D Hair Independent Statistics Consultant7 propensity score matching SPSS V22PS MatchingExample: Comparing 1714 patients with BOTHGout & diabetesto 15,224 patients with ONLY diabetesMario D Hair Independent Statistics Consultant8 CovariatesUnivariate stats: Comparing BOTHGout & diabetes to those with ONLY diabetesMario D Hair Independent Statistics Consultant9 Group N Mean Std.

8 Dev Mean diff (both- type2) /Odds ratio (95% CI) Effect size (d) Total Cholesterol Type 2 only 14332 - * (- ,- ) Gout & type 2 1633 HDL Cholesterol Type 2 only 14971 .38 - * (- ,- ) Gout & type 2 1691 .38 LDL Cholesterol Type 2 only 14274 .87 - * (- ,- ) Gout & type 2 1593 .84 Triglycerides Type 2 only 14906 * ( , ) Gout & type 2 1678 BMI at risk Type 2 only 14670 * ( , ) Gout & type 2 1652 * p < using t to d conversions d = 2t/sqrt(df) & d = ln(OR)*( 3/ ) Group N Mean Std.

9 Dev Mean diff (both- type2) /Odds ratio (95% CI) Effect size (d) Age Type 2 only 15224 * ( , ) Gout & type 2 1714 Gender (%Male) Type 2 only 15224 * ( , ) Gout & type 2 1714 Smoker (%Current) Type 2 only 15224 * ( , ) Gout & type 2 1714 Thiazide Type 2 only 15224 * ( , ) Gout & type 2 1714 Diuretic Type 2 only 15224 * ( , ) Gout & type 2 1714 PS matching : Using a file with only the covariatesMario D Hair Independent Statistics Consultant10 Warning:PS matching will not work if there are missing values on any variable propensity score matching SPSS V22PS MatchingMario D Hair Independent Statistics Consultant11 However propensity score matching does work if there are missing values on any variable PS matching has more options & diagnosticsMario D Hair Independent Statistics Consultant12PS matching Outputs.

10 DatasetsMario D Hair Independent Statistics Consultant13 Paired cases wide formatMatched cases propensity score matching SPSS V22 OutputMario D Hair Independent Statistics Consultant14PS matching Outputs : DiagnosticsMario D Hair Independent Statistics Consultant15 Control Treated All 15224 1714 Matched 1714 1714 Unmatched 13510 0 Discarded 0 0 Samples sizes of matched dataOverall balance test (Hansen & Bowers, 2010) balance testsRelative multivariate imbalance L1 (Iacus, King, & Porro, 2010)Before matchingAfter matchingMultivariate imbalance measure TreatedMeans ControlSD ControlStd.