Usefulness of administrative databases for epidemiological evaluations and healthcare planning

Decision makers of healthcare organizations need information about efficacy and costs health services. Randomised controlled trials (RCTs) remain the accepted “gold standard” for determining the efficacy of new drugs or medical procedures. Randomised trials alone, however, cannot provide all the relevant information that decision makers need to weighing the implications of particular policies affecting medical therapies. Moreover quality organizations and professional societies need information about applicability of trial findings to the settings and patients of interest. Research using disease and intervention registries, outcome studies using administrative databases and performance indicators adopted by quality improvement methods can all shed light on who is most likely to benefit, what the important tradeoffs are and how policy makers might promote the safe, effective and appropriate use of new interventions. Administrative healthcare databases play today a central role in epidemiological evaluation because of their widespread diffusion and low cost of information. Public healthcare regulatory organizations can assist decision makers in providing information based on available electronic health records, promoting the development and the implementation of the methodological tools suitable for the analysis of administrative databases and answering questions oriented to disease management. The aim of this kind of evaluation is to estimate adherence to best practice (in the setting of evidence based medicine) and potential benefits and harms of specific health policies. Two kinds of evaluations can be performed: 1) spot analysis of data about specific questions: geographical distribution of medical procedures like PCI, patterns of transfers for myocardial infarction (hub and spoke model in the emergency network), continuity of care in drug prescription; analysis of data needs a scientific approach: a specific question must be stated (hypothesis); evidence must be searched in the data, in order to verify or to refute the hypothesis. 2) implementation of disease and intervention registries based on administrative databases in order to monitor the diffusion of new procedures or the effects health policy interventions. There is an increasing agreement among epidemiologists on validity of disease and intervention registries based on administrative databases. The study population is composed by the whole resident population. The suitable electronic medical records for analysis are: hospital discharge data, specialist ambulatory records, drug prescription data. Complementary information can be obtained to cope with specific questions (“file protesi”, “archivio esenzioni”, “anagrafe assistiti”): The records of all these files may be linked in order to achieve the correct information about the basic observation unit: the individual patient/subject. Deterministic record linkage is the preferred tool, given the good quality of such data. The most critical issue for observational studies using administrative databases is represented by selection criteria of the observation units: several different criteria may be used, depicting different images of prevalence or incidence of diseases and it is mandatory to perform sensitivity analysis in order to evaluate validity of the estimates. In outcome research in observational studies appropriate risk adjustment tools are needed. Hospital discharge records may be analysed with the indicators developed by AHRQ (Agency for Healthcare Research and Quality) that include efficient risk adjustment tools within a multiple logistic regression model. In disease management programs the Johns Hopkins ACG methodology and the CRG classification system have been proposed, but the applicability to the Italian health information system has to be established. Statistical analysis can be performed by means of multiple logistic regression model for outcome research and by means of survival analysis in order to study failure times (hospital readmissions, continuity of drug prescriptions, survival times). The evidence obtained by means of administrative databases is the first level evidence (a screening tool) useful for targeting areas for which further investigation is needed; the evaluation has to be integrated by second level indicators (performance indicators obtained form clinical medical records). Information can be extracted by means of human work on a sample of medical records or extensive information retrieval can be performed by means of text mining techniques. Textual healthcare databases will represent in the next future a new milestone in epidemiological evaluation