Objectives: To describe and test a model that compares the accuracy of data gathered prospectively versus retrospectively among adult emergency department patients admitted with chest pain.
Methods: The authors developed a model of information flow from subject to medical record to the clinical study case report form, based on a literature review. To test this model, a bidirectional (prospective and retrospective) study was conducted, enrolling all eligible adult patients who were admitted with a chief complaint of chest pain. The authors interviewed patients in the emergency department to determine their chest pain history and established a prospective database; this was considered the criterion standard. Then, patient medical records were reviewed to determine the accuracy and completeness of the information available through a retrospective medical record review.
Results: The model described applies the concepts of reliability and validity to information passed on by the study subject, the clinician, and the medical record abstractor. This study was comprised of 104 subjects, of which 63% were men and the median age was 63 years. Subjects were uncertain of responses for 0-8% of questions and responded differently upon reinterview for subsets of questions 0-30% of the time. The sensitivity of the medical record for risk factors for coronary artery disease was 0.77 to 0.93. Among the 88 subjects (85%) who indicated that their chest pain was substernal or left chest, the medical record described this location in 44%. Timing of the chest pain was the most difficult item to accurately capture from the medical record.
Conclusions: Information obtained retrospectively from the abstraction of medical records is measurably less accurate than information obtained prospectively from research subjects. For certain items, more than half of the information is not available. This loss of information is related to the data types included in the study and by the assumptions that a researcher performing a retrospective study makes about implied versus explicitly stated responses. A model of information flow that incorporates the concepts of reliability and validity can be used to measure some of the loss of information that occurs at each step along the way from subject to clinician to medical record abstractor.