As part of a large collaborative effort, scientists are using decision theory to devise prediction models for coronary artery disease with patient records and medical information stored in the electronic health records. Improving diagnosis at the primary care level will lead to a more favourable clinical outcome.
Clinical practice, especially in the primary care setting, entails many decision-making processes. The process itself is like solving a puzzle, where the correct pieces are required to make sense of the larger picture. The puzzle analogy relates to the series of questions clinicians ask to decipher and piece together a diagnosis, to obtain a favourable outcome depending on the history of the patient. To aid clinicians in these processes, mathematical tools called clinical prediction rules (CPR) are implemented. The process of obtaining a valid CPR includes data mining of patient characteristics and investigation by selecting, exploring and modelling large amounts of data now generated and maintained by various healthcare companies and data servers. One field where CPR is a playing a major role is in the prediction of coronary artery disease (CAD). CAD usually manifests as angina, cardiac arrest, and myocardial infarction with sudden cardiac death, therefore predicting these conditions becomes imperative in the clinical diagnosis. Scientists are now using data from previously published studies to formulate a valid CPR model.
In a study published in the Journal of Clinical Epidemiology, researchers constructed a CPR model for CAD that can be applied to individuals who present symptomatic chest pains at primary care. As part of a larger clinical trial called INTERCHEST, a collaborative effort between many research centres, patient data was evaluated from pooled meta-data sets and key predictors were identified and subsequently combined as one prediction rule. A systematic literature study was performed by searching online medical databases PUBMED and EMBASE, followed by independent reviewers selecting the appropriate studies. After screening roughly 1800 abstracts, 5 studies which fulfilled the inclusion criteria were selected. Studies were excluded if the patients were based at an emergency department or any case of coronary disease was suspected. Subsequently, a meta-analysis was performed to evaluate predictors from medical data for 3,099 patients from five different studies.
The results revealed that six equally weighted predictors could be included in this CPR; namely – age (≥55 (males), ≥65 (females)) (+1), attending physician suspected a serious condition (+1), history of CAD (+1), pain brought on by exertion (+1), pressure-like pain (+1), and pain reproducible palpation (+1). If the total prediction score was <2, CAD was considered absent. When applied to a study set with a known CAD prevalence of 13.2%, the formulated CPR yielded a probability of 2.1% at a score of <2, and 43% at a score of ≥2, within the allotted confidence intervals. Although the present CPR could not be compared with the previously established predictor rules set in specialty clinics and hospitals, the study suggests a more general model for diagnostic research. More importantly, sharing of such clinical data at the primary care level improves the care of the patients and diagnosis of severe coronary conditions. With the increasing dependency diagnostic tools, clinicians have the choice to focus on individualized care particularly for complex diseases such as CAD.
Written By: Akshita Wason, B. Tech, PhD