Improving Quality by Maintaining Accurate Problems in the EHR (IQ-MAPLE)

The overall goal of the IQ-MAPLE project is to improve the quality of care provided to patients with several heart, lung and blood conditions by facilitating more accurate and complete problem list documentation. In the first aim, the investigators will design and validate a series of problem inference algorithms, using rule-based techniques on structured data in the electronic health record (EHR) and natural language processing on unstructured data. Both of these techniques will yield candidate problems that the patient is likely to have, and the results will be integrated. In Aim 2, the investigators will design clinical decision support interventions in the EHRs of the four study sites to alert physicians when a candidate problem is detected that is missing from the patient's problem list - the clinician will then be able to accept the alert and add the problem, override the alert, or ignore it entirely. In Aim 3, the investigators will conduct a randomized trial and evaluate the effect of the problem list alert on three endpoints: alert acceptance, problem list addition rate and clinical quality.

The clinical problem list is a cornerstone of the problem-oriented medical record. Problem lists are used in a variety of ways throughout the process of clinical care. In addition to its use by clinicians, the problem list is also critical for decision support and quality measurement. Patients with gaps in their problem list face significant risks. For example, if a hypothetical patient has diabetes properly documented, his clinician would receive appropriate alerts and reminders to guide care. Additionally, the patient might be included in special care management programs and the quality of care provided to him would be measured and tracked. Without diabetes on his problem list, he might receive none of these benefits. In this study, the investigators developed an clinical decision support intervention that will identify patients with problem lists gaps. The investigators will alert providers of these likely gaps and offer providers the opportunity to correct them. In the first aim, the investigators will design and validate a series of problem inference algorithms, using rule-based techniques on structured data in the electronic health record (EHR) and natural language processing on unstructured data. Both of these techniques will yield candidate problems that the patient is likely to have, and the results will be integrated. In Aim 2, the investigators will design clinical decision support interventions in the EHRs of the four study sites to alert physicians when a candidate problem is detected that is missing from the patient's problem list - the clinician will then be able to accept the alert and add the problem, override the alert, or ignore it entirely. In Aim 3, the investigators will conduct a randomized trial and evaluate the effect of the problem list alert on three endpoints: alert acceptance, problem list addition rate and clinical quality.

Asthma, Atrial Fibrillation, Chronic Obstructive Pulmonary Disease, Coronary Artery Disease, Congestive Heart Failure, Hyperlipidemia, Hypertension, Myocardial Infarction, Sickle Cell Disease, Sleep Apnea, Smoking, Stroke, Tuberculosis

Sites will configure their EHR systems so that alerts will not be triggered for providers in the control arm if the patient does not have the condition on her/his problem list.

Sites will configure their EHR systems so that alerts for these conditions will be triggered for providers in the intervention arm if the patient does not have the condition on her/his problem list. Each alert will be actionable and allow the provider to add the problem to her or his patient's problem list with a single click. The provider will also be able to override the rule of the patient does not have the condition (in which case the alert will not be displayed again unless new information that would trigger the alert is added to the patient's record), or defer the alert until later.

Sites will configure their EHR systems so that alerts for these conditions will be triggered for providers in the intervention arm if the patient does not have the condition on her/his problem list. each alert will be actionable and allow the provider to add the problem to her or his patient's problem list with a single click. The provider will also be able to override the rule of the patient does not have the condition (in which case the alert will not be displayed again unless new information that would trigger the alert is added to the patient's record), or defer the alert until later.

Measuring the rate of acceptance of alerts calculated by number of acceptances for each alert divided by the total number of unique presentations of the alert

Acceptance of the alerts: This first endpoint is descriptive: the acceptance rate for the alerts presented to providers. This will be calculated by taking the total number of acceptances for each alert and dividing it by the total number or unique presentations of the alert. We will conduct a stratified analysis to look at differences in acceptance rates by institution, specialty, disease and provider demographic characteristics, and will report the results in tabular form.

Determining the effect of problem list completion by comparing the number of study-related problems added to problem lists in the electronic health record

Effect on the rate of problem list completion: In this endpoint, we will compare the number of study-related problems added to patient problems lists in the electronic health record in the intervention and control groups.

Determining the quality of care impact of adding suggested problems to the problem list based on 4 outcome measures from NCQA's HEDIS 2013 measure set

Effect on quality of care: Because a key goal of our study is improving clinical outcomes, we have selected four outcome measures to evaluate from NCQA's Healthcare Effectiveness Data and Information Set (HEDIS) 2013 measure set: LDL control in patients with a history of myocardial infarction, LDL control in patients with coronary artery disease, blood pressure control in patients with coronary artery disease and blood pressure control in patients with hypertension. The details for the numerator and denominator for each measure are given in the HEDIS manuals, and our study team will employ NCQA's procedures for calculation of each measure, with modifications as needed given the clinical nature of our dataset.

Evaluating process measures using key process measures for each study condition from CMS, NHLBI, and NQMC

Improvements for process measures To complete the clinical endpoints in the third outcome, we will also evaluate process measures, specifically frequency of LDL testing, prescription of antihyperlipidemic agents, prescription of aspirin or other antiplatelet agents and prescription of antihypertensive agents. We will analyze the results using logistic regression with fixed effects for intervention group (versus control) and site and estimation of the regression parameters with generalized estimating equations (GEE), accounting for clustering between the patients in the same physician as well as patients with different physicians in the same matched pair. We will build separate regression models for each quality measure, and also conduct a pooled analysis with additional effects for quality measure and availability of CDS for the associated measure at the site, in order to estimate the extent to which IQ-MAPLE's effect on quality is mediated by CDS.

Inclusion Criteria: All providers over the age of 18 that use the electronic health record at the specific site that the intervention is being observed. Exclusion Criteria: -

Wright A, Goldberg H, Hongsermeier T, Middleton B. A description and functional taxonomy of rule-based decision support content at a large integrated delivery network. J Am Med Inform Assoc. 2007 Jul-Aug;14(4):489-96. doi: 10.1197/jamia.M2364. Epub 2007 Apr 25.

Kaplan DM. Clear writing, clear thinking and the disappearing art of the problem list. J Hosp Med. 2007 Jul;2(4):199-202. doi: 10.1002/jhm.242. No abstract available.

Szeto HC, Coleman RK, Gholami P, Hoffman BB, Goldstein MK. Accuracy of computerized outpatient diagnoses in a Veterans Affairs general medicine clinic. Am J Manag Care. 2002 Jan;8(1):37-43.

Tang PC, LaRosa MP, Gorden SM. Use of computer-based records, completeness of documentation, and appropriateness of documented clinical decisions. J Am Med Inform Assoc. 1999 May-Jun;6(3):245-51. doi: 10.1136/jamia.1999.0060245.

Carpenter JD, Gorman PN. Using medication list--problem list mismatches as markers of potential error. Proc AMIA Symp. 2002:106-10.

Hartung DM, Hunt J, Siemienczuk J, Miller H, Touchette DR. Clinical implications of an accurate problem list on heart failure treatment. J Gen Intern Med. 2005 Feb;20(2):143-7. doi: 10.1111/j.1525-1497.2005.40206.x.

Wright A, Chen ES, Maloney FL. An automated technique for identifying associations between medications, laboratory results and problems. J Biomed Inform. 2010 Dec;43(6):891-901. doi: 10.1016/j.jbi.2010.09.009. Epub 2010 Sep 25.

Wright A, Pang J, Feblowitz JC, Maloney FL, Wilcox AR, Ramelson HZ, Schneider LI, Bates DW. A method and knowledge base for automated inference of patient problems from structured data in an electronic medical record. J Am Med Inform Assoc. 2011 Nov-Dec;18(6):859-67. doi: 10.1136/amiajnl-2011-000121. Epub 2011 May 25.