The PSMATCH Procedure - SAS

1 SAS/STAT User s GuideThe PSMATCH ProcedureThis document is an individual chapter fromSAS/STAT User s correct bibliographic citation for this manual is as follows: SAS Institute Inc. User s guide . Cary, NC:SAS Institute User s GuideCopyright 2016, SAS Institute Inc., Cary, NC, USAAll Rights Reserved. Produced in the United States of a hard-copy book: No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or byany means, electronic, mechanical, photocopying, or otherwise, without the prior written permission of the publisher, SAS a web download or e-book: Your use of this publication shall be governed by the terms established by the vendor at the timeyou acquire this scanning, uploading, and distribution of this book via the Internet or any other means without the permission of the publisher isillegal and punishable by law.

2 Please purchase only authorized electronic editions and do not participate in or encourage electronicpiracy of copyrighted materials. Your support of others rights is Government License Rights; Restricted Rights:The Software and its documentation is commercial computer softwaredeveloped at private expense and is provided with RESTRICTED RIGHTS to the United States Government. Use, duplication, ordisclosure of the Software by the United States Government is subject to the license terms of this Agreement pursuant to, asapplicable, FAR , DFAR (a), DFAR (a), and DFAR , and, to the extent required under law, the minimum restricted rights as set out in FAR (DEC 2007). If FAR is applicable, this provisionserves as notice under clause (c) thereof and no other notice is required to be affixed to the Software or documentation.

3 TheGovernment s rights in Software and documentation shall be only those set forth in this Institute Inc., SAS Campus Drive, Cary, NC 27513-2414 November 2016 SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in theUSA and other countries. indicates USA brand and product names are trademarks of their respective software may be provided with certain third-party software, including but not limited to open-source software, which islicensed under its applicable third-party software license agreement. For license information about third-party software distributedwith SAS software, refer 95 The PSMATCH ProcedureContentsOverview: PSMATCH Procedure ..7676 Process of Propensity Score Analysis ..7677 Features of the PSMATCH Procedure .

4 7679 Getting Started: PSMATCH Procedure ..7680 Syntax: PSMATCH Procedure ..7688 PROC PSMATCH Statement ..7688 ASSESS Statement ..7691BY Statement ..7694 CLASS Statement ..7695 FREQ Statement ..7695 MATCH Statement ..7695 OUTPUT Statement ..7701 PSDATA Statement ..7702 PSMODEL Statement ..7702 STRATA Statement ..7703 Details: PSMATCH Procedure ..7704 Observational Studies Contrasted with Randomized Trials ..7704 Propensity Score Analysis ..7705 Propensity Score Weighting ..7707 Propensity Score Stratification ..7709 Matching Process ..7709 Matching Statistics ..7711 Matching Methods ..7712 Variable Balance Assessment ..7713 Table Output ..7716 ODS Table Names ..7717 Graphics Output ..7718 ODS Graphics ..7719 Examples: PSMATCH Procedure ..7720 Example : Propensity Score Weighting.

5 7721 Example : Propensity Score Stratification ..7728 Example : Optimal Variable Ratio Matching ..7737 Example : Greedy Nearest Neighbor Matching ..7742 Example : Matching with Replacement ..7751 Example : Mahalanobis Distance Matching ..7756 Example : Matching with Existing Propensity Scores in the Input Data Set ..7761 References ..77667676 FChapter 95: The PSMATCH ProcedureOverview: PSMATCH ProcedureIn a randomized study, such as a randomized controlled trial, the subjects are randomly assigned to a treated(exposure) group or a control (non-exposure) group. Random assignment ensures that the distribution of thecovariates is the same in both groups, and the treatment effect can be estimated by directly comparing theoutcomes for the subjects in the two contrast, the subjects in an observational study, such as a retrospective cohort study or a nonrandomizedclinical trial, are not randomly assigned to the treated and control groups.

6 Confounding can occur ifsome covariates are related to both the treatment assignment and the outcome. Consequently, there can besystematic differences between the treated subjects and the control subjects. In the presence of confounding,statistical approaches are required that remove the effects of confounding when estimating the effect such approach is regression adjustment, which estimates the treatment effect after adjusting for differ-ences in the baseline covariates. However, this approach has practical limitations, as discussed by Austin(2011a). Propensity score analysis is an alternative approach that circumvents many of these propensity score was defined by Rosenbaum and Rubin (1983, p. 47) as the probability of assignmentto treatment conditional on a set of observed baseline covariates. Propensity score analysis minimizes theeffects of confounding and provides some of the advantages of a randomized study.

7 The basis for propensityscore methods is the causal effect model introduced by Rubin (1974).The PSMATCH Procedure provides a variety of tools for propensity score analysis. The Procedure eithercomputes propensity scores or reads previously-computed propensity scores, and it provides the followingmethods for using the scores to allow for valid estimation of treatment effect in a subsequent outcomeanalysis: Inverse probability of treatment weighting and ATT weighting (weighting by odds): The procedurecomputes weights from the propensity scores. These weights can then be incorporated into a subsequentanalysis that estimates the effect of treatment. Stratification: The Procedure creates strata of observations that have similar propensity scores. In asubsequent analysis, the treatment effect can be estimated within each stratum, and the estimates canbe combined across strata.

8 Matching: The Procedure matches each treated unit with one or more control units that have a similarvalue of the propensity score. In a subsequent analysis, the treatment effect can be estimated bycomparing outcomes between treated and control subjects in the matched sample. If the outcomevalues for a study are not available prior to matching, only the matched units are needed for , the cost of the trial is reduced (Stuart 2010, p. 2).The PSMATCH Procedure also provides methods for assessing the balance of baseline covariates and othervariables in the treated and control groups after matching, weighting, or stratification. The Procedure itselfdoes not carry out the outcome analysis, nor does it make use of the outcome adequate variable balance has been achieved (as described in the section Process of Propensity ScoreAnalysis on page 7677), and assuming that no other confounding variables are associated with both thetreatment assignment and the outcome, the output data set that is created by the PSMATCH Procedure servesas input for an appropriate statistical Procedure for the outcome of Propensity Score AnalysisF7677 Process of Propensity Score AnalysisA propensity score analysis usually involves the following steps (Guo and Fraser 2015, p.)

9 131) specify a set of confounding variables that might be related to both the treatment assignment andthe use this set of variables to fit a logistic regression model and compute propensity scores. Theresponse is the probability of assignment to the treatment you are using weighting, you compute observation weights for estimating the treatment effect in aweighted outcome you are using stratification or matching, you specify the support region of observations. Observationsoutside this region are not included in the stratification or you are using stratification, you specify the number of strata and create strata of observations thathave similar propensity you are using matching, you specify criteria such as the matching statistic (the distance metric forcomparing the similarity of subjects) and the method for creating matched sets of observations.

10 Youcan also compute weights for matched assess the balance of variables by comparing the distributions between the treated and improve the balance, you can repeat the process with a different set of variables for the logisticregression model, a different region of support for stratification and matching, a different set ofmatching criteria, or a different matching you are satisfied with the variable balance, you save the output data set for subsequent that the outcome variable is intentionally not used in this process, and the variable selection is notrelated to the observed outcomes (Rubin 2001; Stuart 2010, p. 5). Any variables that might have beenaffected by the treatment should not be included in the process (Rosenbaum and Rubin 1984; Stuart 2010, ).The flowchart in Figure summarizes these 95: The PSMATCH ProcedureFigure Score AnalysisAfter balance is achieved, you can add the response variable to the output data set that PROC PSMATCH created and perform an outcome analysis that mimics the analysis you would perform with data from arandomized study.