Assessing the accuracy of an inter-institutional automated patient-specific health problem list

Abstract

BackgroundHealth problem lists are a key component of electronic health records and are instrumental in the development of decision-support systems that encourage best practices and optimal patient safety. Most health problem lists require initial clinical information to be entered manually and few integrate information across care providers and institutions. This study assesses the accuracy of a novel approach to create an inter-institutional automated health problem list in a computerized medical record (MOXXI) that integrates three sources of information for an individual patient: diagnostic codes from medical services claims from all treating physicians, therapeutic indications from electronic prescriptions, and single-indication drugs.MethodsData for this study were obtained from 121 general practitioners and all medical services provided for 22,248 of their patients. At the opening of a patient's file, all health problems detected through medical service utilization or single-indication drug use were flagged to the physician in the MOXXI system. Each new arising health problem were presented as 'potential' and physicians were prompted to specify if the health problem was valid (Y) or not (N) or if they preferred to reassess its validity at a later time.ResultsA total of 263,527 health problems, representing 891 unique problems, were identified for the group of 22,248 patients. Medical services claims contributed to the majority of problems identified (77%), followed by therapeutic indications from electronic prescriptions (14%), and single-indication drugs (9%). Physicians actively chose to assess 41.7% (n = 106,950) of health problems. Overall, 73% of the problems assessed were considered valid; 42% originated from medical service diagnostic codes, 11% from single indication drugs, and 47% from prescription indications. Twelve percent of problems identified through other treating physicians were considered valid compared to 28% identified through study physician claims.ConclusionAutomation of an inter-institutional problem list added over half of all validated problems to the health problem list of which 12% were generated by conditions treated by other physicians. Automating the integration of existing information sources provides timely access to accurate and relevant health problem information. It may also accelerate the uptake and use of electronic medical record systems.

Restoring Mitochondrial Function: A Small Molecule-mediated Approach to Enhance Glucose Stimulated Insulin Secretion in Cholesterol Accumulated Pancreatic beta cells

“…Support Vector Machines developed by Cortes and Vapnik et al . 17 has recently been employed extensively in biological prediction problem 18 , clinical decision making 19 & risk prediction of common diseases like prediabetes and diabetes 20 . To provide a background of the underlying theory, such problems are typically modelled as quadratic optimization problems.…”

A Quality Improvement Study to Improve Inpatient Problem List Use

“…8 Several studies have been conducted concerning problem list use. Most studies to date have focused on standardizing language within the problem list, 9 natural language processing as a means of capturing problems written by the physician in a note to import to the problem list, 10-12 forcing physicians to enter a diagnosis for which medications were ordered in the inpatient setting to populate the problem list, 13,14 or building alerts around diagnostic rules that notify physicians to add particular problems to the list as a means of generating a problem list. 15 To the best of our knowledge, no studies have been conducted on improving use of the problem list itself without reliance on other techniques to automatically add problems to the problem list.…”

Functional evaluation of out-of-the-box text-mining tools for data-mining tasks

“… 12–16 Raw clinical text arguably provides the most complete picture of the state of patients at any point in time since much of the structured data in EHRs, such as administrative codes, are used primarily for purposes other than communication of key clinical information about patients, for example, billing. 17–20 However, clinical text is unstructured data, and basic questions that are easy to state in plain language are often difficult to reduce to practice, for example, find all patients who have peripheral artery disease (PAD) and who are taking cilostazol. A critical first step in the use of clinical text to address such electronic phenotyping problems is finding mentions of entities of interest, such as drugs, diseases, or laboratory values, in the text.…”