Nonresponse bias in a survey of patient perceptions of hospital care

Background: Incomplete participation is of particular concern for surveys of patient perceptions of care because patients who have negative opinions may be least likely to participate. Objective: We sought to examine indirect evidence of nonresponse bias. Design: We re-analyzed data from a cross-sectional patient survey. Subjects: Our subjects were patients discharged from a Swiss hospital (n = 2156). Measures: We measured the following: (1) an observed problem score, based on 15 key items of the Picker Patient Experience questionnaire, (2) a predicted problem score, and (3) a participation propensity score. The latter scores were computed for all eligible patients, including those who did not return the survey, from routinely available baseline data. Results: The participation rate was 70% (n = 1518), and the mean problem score was 29.9 (SD 23.8). Early respondents reported significantly fewer problems than late respondents (28.6 versus 32.9, P = 0.001). Participation propensity scores were progressively lower in early respondents (mean 74.2), late respondents (70.7), and nonrespondents (63.9, P < 0.001); the pattern was similar for predicted problem scores (early respondents: 29.5; late respondents: 30.5; nonrespondents: 33.4, P < 0.001). The propensity to participate was negatively associated with the problem score (Pearson r = -0.19). Finally, predictors of participation were similar to predictors of problem scores. Conclusions: The tendency to participate in the survey was negatively associated with the report of problems during hospitalization. Nevertheless, increasing participation from 30% to 70% had only a modest influence on the final conclusions of the survey.