The trial period was from July 1, 2003 through February 1, 2005. The sample included 8 community-based and 6 hospital-based primary care practices affiliated with a large urban academic medical center; these practices were stratified by size and type (women’s health versus general primary care and community health center [CHC] versus non-CHC). We then randomized half of these clinics to have their physicians receive CDS for each hypertensive patient as an intervention and half to provide usual care without computerized decision support. In addition to our trial of CDS, 1 hospital-based clinic from each of the study arms was randomly selected to participate in a concurrent pilot trial examining the effectiveness of interval visits for hypertensive patients with a specialized nurse practitioner (NP) between regularly scheduled primary care appointments. Figure illustrates the clinic assignment and patient selection used for the study.
Figure 1 Flow diagram of clinic randomization and nurse practitioner subject recruitment. We analyzed a total of 2,027 patients receiving care in clinics assigned to either computerized decision support (CDS) for all providers within the practice or no CDS. One (more ...)
We examined data obtained from the electronic medical record (EMR) of patients older than 20 years with a diagnosis of hypertension to determine whether the use of CDS improved physician prescribing of guideline-recommended drug therapy and levels of blood pressure control and if the effects of this intervention differed by patient race/ethnicity. The Human Studies Committee at the Brigham and Women’s Hospital approved the study protocol.
Study Sites and Patient Sample
We studied adult patients with at least 1 hypertension-related outpatient visit to 1 of the study clinics during the 1-year period before the beginning of the interventions. From these patients, we examined records of patients aged >20 years, who had at least 2 hypertension-related outpatient visits to 1 of the participating practices during the intervention trial, and whose race/ethnicity was available (96%) in administrative data. All of the clinics involved in the study used EMR, including electronic prescribing of medications, for each patient during the study period. To identify hypertension-related visits, we reviewed the electronic medical record for all clinic visits with a primary or secondary diagnostic code of hypertension (HTN) (ICD-9 401–401.9, 405–405.99).
Using national guidelines or standards including the Health Plan Employer Data and Information Set (HEDIS), the sixth and seventh reports of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure (JNC VI and VII), and the American Heart Association/American College of Cardiology (AHA/ACC) 2001 guidelines for cardiovascular disease prevention, we developed racial and disease-specific (i.e., diabetes, congestive heart failure, coronary heart disease) algorithms for blood pressure treatment emphasizing the pharmacologic drug class recommended for each condition.1,6,25,26
Computerized Decision Support Protocol
We used similar methods to our previously described study of electronic decision support.18
Each time that a clinician opened a patient’s chart, an algorithm was run within the EMR to determine whether the patient was receiving a medication in an antihypertensive drug class in accordance with JNC guidelines. The algorithm searched patients’ vital signs, problem lists, medication lists, and allergy lists within the electronic record. CDS was automatically generated for a patient based on a documentation of hypertension in the problem list, or an average of the patient’s 3 most recent documented blood pressure readings ≥140/90 for patients without an existing diagnosis of hypertension, and the patient not having any medications in the guideline-recommended drug class on their medication list. Patients with a listed allergy to a drug in the recommended class were not considered eligible to receive the recommended therapy. The reminders were extensively tested before implementation using dummy patients within the EMR.
Before the start of the intervention, participating physicians were sent a global e-mail notifying them of new hypertension-related reminders. Once the CDS was generated, all clinicians practicing within clinics randomized to CDS received the appropriate reminders. Reminders were displayed within the EMR on the main patient summary screen when the patient’s record was accessed; these reminders were suppressed for physicians practicing in clinics in the control arm. In addition to the electronic display, the reminders generated from the CDS could also be printed off onto a paper version generated using the same algorithms as the electronic reminders; these reminders could be printed on a patient summary page and were distributed to physicians at the beginning of a practice session. Table demonstrates examples of the reminders generated by our CDS for physicians. Algorithms for each of the CDS rules and subsequent reminders are available in the Appendix
Samples of Computerized Medical Reminders and Case Management Duties
Nurse Practitioner Protocol
During the pilot NP trial, patients reporting for routine hypertension care to a clinic assigned to the NP arm were called by study staff within 1 week of their most recent visit during the study period to explain the purpose of the NP and to obtain their verbal consent to see the NP at their usual source of care within 2 weeks of that most recent physician visit. Consented patients met with a research assistant in the waiting room at the time of their scheduled NP appointments and were escorted to the NP examination room located in the same clinic. Table lists the NP duties performed during each visit. One NP conducted all patient visits and follow-up phone calls.
Providers practicing in clinics randomized to the NP had the option of refusing patients for NP co-management. Of the 1,452 patients receiving care in NP clinics during the study, 921 (63%) were consented and had at least 1 visit to the NP; of these, only 193 (21%) met inclusion criteria for our analyses (Fig. ).
Medical Record Review
Through an electronic query of each participant’s EMR we obtained: patient race/ethnicity, sex, age at the time of the first visit, primary insurer, drug allergies, comorbid diseases (diabetes, congestive heart failure [CHF], coronary artery disease [CAD], or renal failure) as listed on the patient problem list, and classes of antihypertensive drugs on the medication list. In addition, we examined electronic prescribing records for each patient, collecting data on new antihypertensive medication prescribing within 7 days of each reminder date. We selected a period of 7 days to identify medication changes that were more likely the direct result of the reminder being seen by the provider.18
Trained abstracters then reviewed the EMR for each patient’s blood pressure for the first and last hypertension-related visits during the study period to obtain participants’ index and outcome visit blood pressures. We defined the last hypertension-related visit during the study period as the outcome visit to reflect the HEDIS measure of most recent blood pressure control and to allow the most time between physician exposure to CDS and clinical outcomes. Abstracters obtained blood pressure readings from 2 sources: (1) the vital signs field of the electronic medical record, which contains blood pressure measures typically entered by a nurse or medical assistant when the patient registers for an appointment, and (2) the text of the provider’s encounter note for each visit. We examined the text of providers’ notes for blood pressure readings because nurses/medical assistants did not uniformly transfer readings into the electronic vital signs field in all clinics until January of 2005 when all the affiliated clinics had electronic blood pressure devices that allowed them to directly download readings into the electronic record.
If more than 1 blood pressure was documented for a visit, we averaged them to obtain the mean systolic and diastolic blood pressure. We also classified each mean blood pressure as controlled (<130/80 for patients with diabetes or renal failure or <140/90 for other patients) or uncontrolled. Visits with undocumented blood pressures were also classified as uncontrolled. We found blood pressure was documented for 97% of index and 95% of outcome visits. For a random subset of 30 records, we tested for interrater reliability and found excellent agreement for our data abstraction instrument among reviewers (kappa
Our sample of 14 clinics was chosen on the basis of feasibility for CDS. After assuming a 20% increase in sample size to account for within-physician correlation, we estimated a minimum of 943 patients per CDS arm would provide 80% power to detect a 10% absolute increase in blood pressure control and guideline compliance rates in the intervention arm compared to the control arm. Before the intervention, we estimated that an 18-month trial would be sufficient to enroll 2,274 eligible patients in the CDS and usual care study arms.
In our analyses, patients are the primary unit of analysis; however, randomization occurred at the level of the clinic and CDS was targeted to providers nested within the clinic. For this reason, all analyses were carried out using the SUDAAN software package to account for correlation between patients seen by the same physician and between physicians practicing within the same clinic.27
We tested the effectiveness of the CDS using “intent-to-treat” analyses; all patients’ whose initial visits occurred in an intervention clinic were assigned to that intervention throughout the analyses. Study endpoints included blood pressure control and mean systolic and diastolic blood pressures at the outcome visit. In addition, for each encounter where a reminder was generated or suppressed we examined whether the provider prescribed a medication in the recommended drug class within 1 week of the visit date. We compared patients’ baseline demographic and clinical characteristics by study arm (Table ) using the Pearson chi-square test for categorical variables and analysis of variance (ANOVA) for continuous variables. We report two-tailed P
values with statistical significance set at P
0.05 for these analyses.
Demographic and Clinical Characteristics of Study Participants by Intervention Arm
We analyzed a series of multivariable logistic regression models to assess whether CDS was associated with the prescribing of a guideline recommended drug class or outcome visit blood pressure control compared to usual care. Using linear regression, we assessed whether the intervention was associated with improved outcome visit mean systolic and diastolic blood pressure. All models adjusted for patients’ sociodemographic characteristics. To adjust for severity of illness, models included an adjustment for the number of patient care visits each patient received the year before the intervention. In addition, because the NP pilot was also aimed at improving blood pressure control, models predicting outcome blood pressure levels also contained an adjustment for NP and patients’ baseline blood pressure. Data were available on every variable for all 2,027 participants for multivariable logistic regression and for 1,879 of the 2,027 participants (93%) for the multivariable linear regressions; participants without complete data were excluded from linear regressions. We report adjusted odds ratios with 95% confidence intervals from logistic models and the adjusted least square means from linear regression for each predictor along with its P value.
We performed several secondary analyses. To determine whether CDS reduced any racial/ethnic differences in measured outcomes, we included interaction terms for patient race/ethnicity and intervention arm. All nonsignificant interaction terms were removed from the final model. To examine any potential misclassification of patients without documented blood pressure, we repeated our analyses excluding patients with no documented index or outcome blood pressure.