MP313IDENTIFYING OUTLYING PRACTICES IN PREVALENCE OF CKD IN PRIMARY CARE

Abstract

Introduction and Aims: Wide variation exists between practices in terms of both testing and coding of chronic kidney disease (CKD). This makes identification of outlying practices for CKD prevalence problematic, with standard funnel plot methods resulting in large numbers of practices flagged as outliers. This work aims to account for some of this heterogeneity (observed as overdispersion in these funnel plots) by adjusting the contours for factors related to the between-practice heterogeneity and using multiplicative random-effects modelling. Winsorisation is also used to reduce the potential influence of extreme values. Methods: Data were collected in 2015 from 915 GP practices in England and Wales as part of the National Chronic Kidney Disease Audit (NCKDA). All patients aged 18+ with either a QOF code for CKD or with a risk factor / renal disease diagnosis for at least one year are included in analysis. Retrospective creatinine and glomerular filtration rate (eGFR) measurements were collected together with baseline characteristics. We aim to identify outlying practices in terms of (i) age-sex standardised prevalence of QOF-coded CKD, (ii) prevalence of uncoded CKD amongst those with eGFR evidence of CKD and (iii) total age-sex standardised CKD prevalence (combining those with coded and uncoded CKD). A series of funnel plots are produced to explore the impact of contour adjustments on identifying outlying practices: 1) Crude contours calculated using standard mean ±2SD / ±3SD 2) Contours adjusted for overdispersion 3) As (2), with 10% winsorisation 4) As (3), with expected number of cases adjusted for key comorbidities (diabetes, hypertension, CVD) 5) As (4), with expected number of cases also adjusted for ethnicity and index of multiple deprivation (IMD) Results: 756 practices have data on list size and all risk factors of interest. The mean age-sex standardised CKD prevalence was 3.1% (SD 1.2%), mean proportion uncoded amongst those with GFR evidence of CKD was 35.8% (SD 17.6%) and mean combined age-sex standardised prevalence was 4.3% (SD 1.2%). The Table shows the number of outlying practices identified from each funnel plot for each of the three outcomes of interest. Conclusions: Random effects modelling of overdispersion in practice CKD prevalence dramatically reduces the number of practices identified as outliers. Further adjustments to reduce the impact of extreme outliers through winsorisation and to account for practice case-mix through adjustment of expected cases have a relatively small impact on outlier identification. (Table presented).