Should measures of patient experience in primary care be adjusted for case mix? Evidence from the English General Practice Patient Survey.

Cambridge Centre for Health Services Research, University of Cambridge, Cambridge, UK.
BMJ quality & safety (Impact Factor: 3.28). 05/2012; 21(8):634-40. DOI: 10.1136/bmjqs-2011-000737
Source: PubMed

ABSTRACT Uncertainties exist about when and how best to adjust performance measures for case mix. Our aims are to quantify the impact of case-mix adjustment on practice-level scores in a national survey of patient experience, to identify why and when it may be useful to adjust for case mix, and to discuss unresolved policy issues regarding the use of case-mix adjustment in performance measurement in health care.
Secondary analysis of the 2009 English General Practice Patient Survey. Responses from 2 163 456 patients registered with 8267 primary care practices. Linear mixed effects models were used with practice included as a random effect and five case-mix variables (gender, age, race/ethnicity, deprivation, and self-reported health) as fixed effects.
Primary outcome was the impact of case-mix adjustment on practice-level means (adjusted minus unadjusted) and changes in practice percentile ranks for questions measuring patient experience in three domains of primary care: access; interpersonal care; anticipatory care planning, and overall satisfaction with primary care services.
Depending on the survey measure selected, case-mix adjustment changed the rank of between 0.4% and 29.8% of practices by more than 10 percentile points. Adjusting for case-mix resulted in large increases in score for a small number of practices and small decreases in score for a larger number of practices. Practices with younger patients, more ethnic minority patients and patients living in more socio-economically deprived areas were more likely to gain from case-mix adjustment. Age and race/ethnicity were the most influential adjustors.
While its effect is modest for most practices, case-mix adjustment corrects significant underestimation of scores for a small proportion of practices serving vulnerable patients and may reduce the risk that providers would 'cream-skim' by not enrolling patients from vulnerable socio-demographic groups.

  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Objectives To determine the extent to which practice level scores mask variation in individual performance between doctors within a practice. Design Analysis of postal survey of patients' experience of face-to-face consultations with individual general practitioners in a stratified quota sample of primary care practices. Setting Twenty five English general practices, selected to include a range of practice scores on doctor-patient communication items in the English national GP Patient Survey. Participants 7721 of 15 172 patients (response rate 50.9%) who consulted with 105 general practitioners in 25 practices between October 2011 and June 2013. Main outcome measure Score on doctor-patient communication items from post-consultation surveys of patients for each participating general practitioner. The amount of variance in each of six outcomes that was attributable to the practices, to the doctors, and to the patients and other residual sources of variation was calculated using hierarchical linear models. Results After control for differences in patients' age, sex, ethnicity, and health status, the proportion of variance in communication scores that was due to differences between doctors (6.4%) was considerably more than that due to practices (1.8%). The findings also suggest that higher performing practices usually contain only higher performing doctors. However, lower performing practices may contain doctors with a wide range of communication scores. Conclusions Aggregating patients' ratings of doctors' communication skills at practice level can mask considerable variation in the performance of individual doctors, particularly in lower performing practices. Practice level surveys may be better used to "screen" for concerns about performance that require an individual level survey. Higher scoring practices are unlikely to include lower scoring doctors. However, lower scoring practices require further investigation at the level of the individual doctor to distinguish higher and lower scoring general practitioners.
    BMJ British medical journal 11/2014; 349:g6034. · 16.30 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: ABSTRACT Aims: This study aims to explore differences between crude and case mix-adjusted estimates of hospital performance with respect to the experience of cancer patients. Materials & methods: This study analyzed the English 2011/2012 Cancer Patient Experience Survey covering all English National Health Service hospitals providing cancer treatment (n = 160). Logistic regression analysis was used to predict hospital performance for each of the 64 evaluative questions, adjusting for age, gender, ethnic group and cancer diagnosis. The degree of reclassification was explored across three categories (bottom 20%, middle 60% and top 20% of hospitals). Results: There was high concordance between crude and adjusted ranks of hospitals (median Kendall's τ = 0.84; interquartile range: 0.82-0.88). Across all questions, a median of 5.0% (eight) of hospitals (interquartile range: 3.8-6.4%; six to ten hospitals) moved out of the extreme performance categories after case mix adjustment. Conclusions: In this context, patient case mix has only a small impact on measured hospital performance for cancer patient experience.
    Future Oncology 12/2013; · 2.61 Impact Factor
  • SEMERGEN - Medicina de Familia 01/2014;

Full-text (2 Sources)

Available from
May 22, 2014