Investigating practice variation in a changing primary care. A multilevel perspective on The Skaraborg Primary Care Database

Abstract: Background and Aims: Differences in the clinical care of patients have been explained through the concept of practice variation, that is, the occurrence of established local medical procedures that varies between geographic areas. The primary measures introduced to eliminate undesirable differences in medical care are the introduction of medical guidelines and economical incentives that rewards the desired behaviour. Practice variation in primary care can be seen on different hierarchically organised levels (e.g. patient, physician, health care center (HCC)) and Multilevel Regression analysis techniques (MLRA) offer a suitable tool to analyse these kinds of data. Data extracted from computerised medical records are ideal for studying practice variation as they often comprise information from several levels. In the region of Skaraborg in Sweden a new database, the Skaraborg Primary care database (SPCD), comprising information extracted form the computerised medical records of all public health care centres has recently been established. The overall aims of this thesis were to examine the usefulness and quality of the SPCD database for research and to study practice variation in some important areas such as diagnosis registration, laboratory analysis ordering and prescriptions. Furthermore, the influence of changes in the economic incentives on physician’s clinical behaviour was investigated. Material and Methods: In all studies data from the SPCD comprising data on individual patients from all public health care centres was used. The registration of diagnoses in the SPCD was validated by comparing the occurrence of recorded diagnosis in the diagnosis register of the database with the free text part of the patient medical records for a randomly selected sample of patients. Multilevel logistic regression analysis was used to investigate practice variation in prescribing and laboratory test ordering, focusing on measures of both frequency and variance. The effects of changes in economic incentives for diagnosis coding and prescribing were examined by comparing multilevel analysis results before and after implementation of the economic incentives. Results and Conclusions: The frequency of registration of ICD codes varied between diagnoses but also between physicians and HCCs. Different diagnoses need to be validated separately. The occurrence of practice variation was demonstrated both in laboratory test ordering where the physician level was the most important level and in prescribing where physician and HCC levels were equally important in explaining the observed variation. A positive effect in adherence to prescribing guidelines was demonstrated after the introduction of a decentralised drug budget. The introduction of a strong economic incentive for ICD coding showed the expected rise in coding rates and decline in variation, directly affecting the diagnoses register of the research database. Changes in the healthcare process will have a direct impact on the research database. Knowledge about the local health care processes is essential when interpreting database data. The SPCD seems as a good complement to previously established databases and quality registers, offering new possibilities when studying primary care.