Methods for work package 2
Work package 2 aimed to assess variation in rates of surgery for SUI using HES APC (inpatient) data. We examined variation in rates of SUI surgery between NHS CCGs and other relevant regional units, and the impact of supply-side factors (e.g. primary care characteristics and availability and organisation of secondary care services) on local rates. WP 2 focused on variation in surgery for SUI specifically because SUI symptoms are the most common subtype of UI symptoms; 50% of women with UI indicate solely SUI symptoms and approximately 40% indicate MUI symptoms, where SUI and UUI symptoms coexist.16,17
The cohort comprised women aged ≥ 20 years who had received surgical treatment for SUI between 1 April 2013 and 31 March 2016 and had a SUI diagnosis recorded at the time of the procedure. SUI surgery was defined using UK OPCS-4 codes (Table 1) based on coding work conducted for WP 1.44 SUI diagnosis was defined using the ICD-10 code N39.3 Stress urinary incontinence.43
The outcome measure was rate of surgery for SUI per 100,000 women per year at two geographic levels: 209 CCGs and 44 Sustainability and Transformation Partnership (STP) areas. CCGs are statutory NHS bodies responsible for the planning and commissioning of health care services in a local area (with an average population size of about 104,000 adult females). CCG areas are grouped into 44 STP areas (with an average population size of about 493,000 adult females), which were set up to co-ordinate improvements in the delivery of NHS services.49 Reference denominator populations were derived by aggregating the 2011 census population counts for women aged ≥ 20 years in lower-layer super output areas (LSOAs) that are within the respective boundaries of the CCG and STP areas. There are 32,844 LSOAs (postcode-based geographic units) in England (with an average population of approximately 1700 people).50 Women may have had repeat procedures in the study period, but only the first procedure was counted in calculating the surgery rate.
Sociodemographic factors may explain variations in the rates of surgery for SUI. We adjusted for age, socioeconomic deprivation, ethnicity and limiting long-term illness in our regression models. We handled age as a patient-level characteristic grouped into five categories (i.e. 20-39, 40-49, 50-59, 60-69 and ≥ 70 years). Socioeconomic deprivation, ethnicity and limiting long-term illness were CCG-level characteristics derived from 2011 census data.50 For socioeconomic deprivation, we used the averages of the national ranking of the Index of Multiple Deprivation (IMD) of LSOAs within each CCG, and grouped the CCG averages into national quintiles ranging from 1 (most deprived CCGs) to 5 (least deprived CCGs).51 For ethnicity, we used the percentage of the population reporting a black, Asian and minority ethnic (BAME) background, and for long-term illness we used the percentage of the population who reported that their day-to-day activities were limited because of a health problem or disability that had lasted, or was expected to last, at least 12 months. For each CCG, we took the averages of these percentages for LSOAs and grouped these CCG averages into national quintiles (range: 1 corresponds to CCGs with average percentages in the lowest quintile to 5 the highest quintile).
We calculated the number and the unadjusted and adjusted rates per 100,000 women per year of SUI procedures overall, and according to patient and regional characteristics. Incidence rate ratios (IRRs) were used to represent associations between the procedure rate and regional characteristics. Multilevel Poisson regression models were used to produce empirical Bayes’ estimates of the unadjusted and adjusted incidence rates for each CCG and STP area. In addition, risk-adjusted regression models were used to assess geographic variation in the rates of surgery by year. The empirical Bayes’ estimator produces more precise results by ‘pulling’ estimates for small outlier regions towards the mean.52 For each geographic area level (CCG/STP), we illustrated the amount of variation in adjusted surgery rates using maps and range plots with 99.8% credibility intervals. CCGs and STPs were marked as ‘outliers’ where the national average rate of surgery was not within the 99.8% credibility interval of their rates. All statistical analyses were performed using Stata, version 15 (StataCorp LP, College Station, TX, USA).