Impact of computerized decision support on blood pressure management and control: a randomized controlled trial

From Clinfowiki
Revision as of 14:42, 28 October 2008 by DeanSittg (Talk | contribs)

Jump to: navigation, search

Review: Computerized Decision Support does not impact indices of blood pressure control

Hicks LS, Sequist TD, Ayanian JZ, Shaykevich S, Fairchild DG, Orav EJ, Bates DW. Impact of computerized decision support on blood pressure management and control: A randomized controlled trial. J Gen Intern Med. 2008; 23(4):429–41

Question Will physicians’ use of computerized decision support (CDS) result in superior hypertension management and outcomes in a racially heterogenous population, relative to patients whose physicians do not use CDS? Also, will CDS help reduce disparities

Subject selection and Independent Variables The subjects were drawn from the patient pool of 8 community based and 6 hospital based outpatient clinics in the Boston area. An initial pool of 5,128 patients were who had at least 1 hypertension related visit to an ambulatory care clinic in the year prior to the experiment were initially targeted. Of this cohort the investigators analyzed the medical records of those patients who satisfied the following criteria: a) over 20 years of age, b) having at least 2 hypertension related visits and c) known ethnicity on the medical record. The doctors treating the patients were randomized to one of two conditions; using CDS or no CDS. Thus, the subjects were not randomly assigned to control or intervention. Instead, the doctors were assigned randomly to the conditions, making the subject assignment pseudo-random. Moreover, the patients were treated at one of 14 outpatient clinics that varied in size and organization (hospital based vs. community health center). Also of interest was the ethnicity of the patient.

Outcomes/Dependent Variables The measures of interest were culled directly from the patient’s electronic medical record. Demographic information, medication history, and co-morbid conditions were recorded. Of particular interest were medication classes that were recommended by the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure (JNC), and whether those drugs were prescribed within a week. Extracted from the medical record were the blood pressure values for the first and last hypertension related visit. The visits were chosen so as to maximize the time between the two visits. Also, the data gatherers classified the blood pressure values as “controlled” or “uncontrolled”.

Main Results To assess the effect of CDS on outcome blood pressures, they used a multivariate linear regression model. The model also included other variables of interest such as sex, insurance type, and ethnicity. Similarly, a multivariate logistic regression model was used to assess the odds ratios of blood pressure control after accounting for these variables. Overall, they found no effect of CDS on blood pressure or blood pressure control. Moreover, CDS did not differentially impact minority patients. CDS did however significantly impact whether the patients were prescribed JNC compliant medications. Moreover, patients with private insurance had higher odds of being prescribed the medication.

Conclusion This study demonstrated the role that EMR’s can play in supporting evidence based medicine. However, it did not show any impact of CDS on outcome measures. It did demonstrate CDSs efficacy for supporting compliance of prescription guidelines. The authors acknowledge some of the limitations of gleaning data from the EMR, especially when they could not ensure patient compliance with the suggested treatments. Also missing from the analysis was the effect of the clinic or clinician. As the authors mention, the group assignments were not perfectly random. As such, the lack of independence on the observations (a cluster of patients being treated by the same doctor) might have introduced some issues in their regression models. On the other hand, it’s possible they determined that these factors did not significantly influence their outcome variable. Yet, its inclusion would be important all the same due to the explicit acknowledgment that it is a confounding variable.

Chia-Hua Yu