BACKGROUND: Randomized trials aimed at improving the quality of medical care often randomize the provider. Such trials are frequently embedded in health care systems with available automated records, which can be used to enhance the design of the trial. METHODS: We consider how available information from automated records can address each of the following concerns in the design of a trial: whether to randomize individual providers or practices; clustering of outcomes among patients in the same practice and its impact on study size; expected heterogeneity in adherence and the response to the intervention; eligibility criteria and the trade-offs between generalizability and internal validity; and blocking or matching to alleviate covariate imbalance across practices. RESULTS: Investigators can use available information from an automated database to estimate the amount of clustering of patients within providers and practices, and these estimates can inform the decision on whether to randomize at the level of the patient, the provider, or the practice. We illustrate calculation of the anticipated design effect for a proposed cluster-randomized trial and its implications for sample size. With available claims data, investigators can apply focused eligibility criteria to exclude subjects and providers with expected low compliance or lower likelihood of benefit, although possibly at some loss of generalizability. Chance imbalances in covariates are more likely when randomization occurs at the level of the practice than at the level of the patient, so we propose a matching score to limit such imbalances by design. CONCLUSIONS: Challenges to compliance, expected small effects, and covariate imbalances are particularly likely in cluster-randomized trials of quality improvement interventions. When such trials are embedded in medical systems with available automated records, use of these data can enhance the design of the trial.
BACKGROUND: Randomized trials aimed at improving the quality of medical care often randomize the provider. Such trials are frequently embedded in health care systems with available automated records, which can be used to enhance the design of the trial. METHODS: We consider how available information from automated records can address each of the following concerns in the design of a trial: whether to randomize individual providers or practices; clustering of outcomes among patients in the same practice and its impact on study size; expected heterogeneity in adherence and the response to the intervention; eligibility criteria and the trade-offs between generalizability and internal validity; and blocking or matching to alleviate covariate imbalance across practices. RESULTS: Investigators can use available information from an automated database to estimate the amount of clustering of patients within providers and practices, and these estimates can inform the decision on whether to randomize at the level of the patient, the provider, or the practice. We illustrate calculation of the anticipated design effect for a proposed cluster-randomized trial and its implications for sample size. With available claims data, investigators can apply focused eligibility criteria to exclude subjects and providers with expected low compliance or lower likelihood of benefit, although possibly at some loss of generalizability. Chance imbalances in covariates are more likely when randomization occurs at the level of the practice than at the level of the patient, so we propose a matching score to limit such imbalances by design. CONCLUSIONS: Challenges to compliance, expected small effects, and covariate imbalances are particularly likely in cluster-randomized trials of quality improvement interventions. When such trials are embedded in medical systems with available automated records, use of these data can enhance the design of the trial.
Authors: Wayne N Welsh; Hsiu-Ju Lin; Roger H Peters; Gerald J Stahler; Wayne E K Lehman; Lynda A R Stein; Laura Monico; Michele Eggers; Sami Abdel-Salam; Joshua C Pierce; Elizabeth Hunt; Colleen Gallagher; Linda K Frisman Journal: Drug Alcohol Depend Date: 2015-04-09 Impact factor: 4.492
Authors: Amanda F Dempsey; Jennifer Pyrznawoski; Steven Lockhart; Juliana Barnard; Elizabeth J Campagna; Kathleen Garrett; Allison Fisher; L Miriam Dickinson; Sean T O'Leary Journal: JAMA Pediatr Date: 2018-05-07 Impact factor: 16.193
Authors: Michael Prendergast; Wayne N Welsh; Lynda Stein; Wayne Lehman; Gerald Melnick; Umme Warda; Michael Shafer; Wendy Ulaszek; Eleni Rodis; Sami Abdel-Salam; Jamieson Duvall Journal: J Behav Health Serv Res Date: 2017-10 Impact factor: 1.505
Authors: Niteesh K Choudhry; Thomas Isaac; Julie C Lauffenburger; Chandrasekar Gopalakrishnan; Nazleen F Khan; Marianne Lee; Amy Vachon; Tanya L Iliadis; Whitney Hollands; Scott Doheny; Sandra Elman; Jacqueline M Kraft; Samrah Naseem; Joshua J Gagne; Cynthia A Jackevicius; Michael A Fischer; Daniel H Solomon; Thomas D Sequist Journal: Am Heart J Date: 2016-08-08 Impact factor: 4.749
Authors: J R Curtis; T Arora; J Xi; A Silver; J J Allison; L Chen; K G Saag; A Schenck; A O Westfall; C Colón-Emeric Journal: Osteoporos Int Date: 2009-03-25 Impact factor: 4.507
Authors: Aimee F English; L Miriam Dickinson; Linda Zittleman; Donald E Nease; Alisha Herrick; John M Westfall; Matthew J Simpson; Douglas H Fernald; Robert L Rhyne; W Perry Dickinson Journal: Ann Fam Med Date: 2018-04 Impact factor: 5.166
Authors: Jason L Vassy; Charles A Brunette; Nilla Majahalme; Sanjay Advani; Lauren MacMullen; Cynthia Hau; Andrew J Zimolzak; Stephen J Miller Journal: Contemp Clin Trials Date: 2018-10-24 Impact factor: 2.226