Detecting referral and selection bias by the anonymous linkage of practice, hospital and clinic data using Secure and Private Record Linkage (SAPREL): case study from the evaluation of the Improved Access to Psychological Therapy (IAPT) service

• Published in: BMC medical informatics and decision making Vol. 11; no. 1; p. 61
• Main Authors: de Lusignan, Simon, Navarro, Rob, Chan, Tom, Parry, Glenys, Dent-Brown, Kim, Kendrick, Tony
• Format: Journal Article
• Language: English
• Published: London BioMed Central 13.10.2011; BMC
• Subjects: Adult; Bias; Clinics; Exports; Female; Health Informatics; Health sciences; Health services; Hospitals; Humans; Information Systems and Communication Service; Male; Management of Computing and Information Systems; Medical Record Linkage - methods; Medicine; Medicine & Public Health; Mental Disorders - rehabilitation; Mental Disorders - therapy; Methods; Middle Aged; Outcome Assessment (Health Care); Primary care; Privacy; Referral and Consultation - statistics & numerical data; Research Article; Selection Bias; Studies; Syntax; Strong Identifier; Common Mental Health Problem; Practice Population; False Negative Count; Electronic Patient Record System
• ISSN: 1472-6947; 1472-6947
• DOI: 10.1186/1472-6947-11-61

Background The evaluation of demonstration sites set up to provide improved access to psychological therapies (IAPT) comprised the study of all people identified as having common mental health problems (CMHP), those referred to the IAPT service, and a sample of attenders studied in-depth. Information technology makes it feasible to link practice, hospital and IAPT clinic data to evaluate the representativeness of these samples. However, researchers do not have permission to browse and link these data without the patients' consent. Objective To demonstrate the use of a mixed deterministic-probabilistic method of secure and private record linkage (SAPREL) - to describe selection bias in subjects chosen for in-depth evaluation. Method We extracted, pseudonymised and used fuzzy logic to link multiple health records without the researcher knowing the patient's identity. The method can be characterised as a three party protocol mainly using deterministic algorithms with dynamic linking strategies; though incorporating some elements of probabilistic linkage. Within the data providers' safe haven we extracted: Demographic data, hospital utilisation and IAPT clinic data; converted post code to index of multiple deprivation (IMD); and identified people with CMHP. We contrasted the age, gender, ethnicity and IMD for the in-depth evaluation sample with people referred to IAPT, use hospital services, and the population as a whole. Results The in IAPT-in-depth group had a mean age of 43.1 years; CI: 41.0 - 45.2 (n = 166); the IAPT-referred 40.2 years; CI: 39.4 - 40.9 (n = 1118); and those with CMHP 43.6 years SEM 0.15. (n = 12210). Whilst around 67% of those with a CMHP were women, compared to 70% of those referred to IAPT, and 75% of those subject to in-depth evaluation (Chi square p < 0.001). The mean IMD score for the in-depth evaluation group was 36.6; CI: 34.2 - 38.9; (n = 166); of those referred to IAPT 38.7; CI: 37.9 - 39.6; (n = 1117); and of people with CMHP 37.6; CI 37.3-37.9; (n = 12143). Conclusions The sample studied in-depth were older, more likely female, and less deprived than people with CMHP, and fewer had recorded ethnic minority status. Anonymous linkage using SAPREL provides insight into the representativeness of a study population and possible adjustment for selection bias.