Electronic Prescription Use by Specialty
Published Online: December 12, 2013
Erik W. J. Kokkonen, MD; Scott A. Davis, MA; Hsien-Chang Lin, PhD; Steven R. Feldman, MD, PhD; and Alan B. Fleischer, Jr, MD
A 2006 Institute of Medicine study noted that there were an estimated 1.5 million preventable adverse drug events (ADEs) per year in the United States, resulting in more than 7000 deaths and costs exceeding $3.5 billion each year.1 A 2005 study estimated that each preventable ADE costs $2000.2 In the ambulatory setting alone, ADEs result in an estimated $887 million annual expense to Medicare—not including patient suffering or loss of productivity.2 In an effort to address these problems, the federal government passed the Medicare Improvements for Patients and Providers Act of 2008, which established a 5-year incentive program for medical professionals to electronically prescribe medications.3
Electronic prescribing (e-prescribing) may improve quality of care and reduce costs by enhancing appropriate drug usage, decreasing the time to renew medications, providing copay information and alternatives to formulary drugs, and reducing ADEs.4,5 It is estimated that 28% to 95% of ADEs may be prevented by electronic prescriptions (e-prescriptions) and computer monitoring systems.6 A separate report found that thousands of dollars per office can be saved by reducing time spent resolving formulary issues or clarifying prescriptions, tasks that may be aided by e-prescribing.7
Although incentives are in place and there is potential to save both lives and costs, the uptake of e-prescribing by the medical community is not well characterized. The purpose of this study was to examine the trends and compare rates of e-prescriptions from 2007 to 2010 among 14 different medical specialties.
Nationally representative data from the National Ambulatory Medical Care Survey (NAMCS) were analyzed from 2007 to 2010. The National Ambulatory Medical Care Survey uses a multistage probability design. First, a random sample of practicing physicians in the United States is selected and then office visits during a 7-day period from these physicians’ practices are sampled. Although the
number of physicians sampled in more than 1 year cannot be determined, NAMCS randomly sampled about 0.69% of all eligible physicians in 2010, suggesting that the rate of repeat sampling in any 2 years is most likely less than 1%. The NAMCS data include symptoms, diagnoses, services provided, medications prescribed, patient demographics, and e-prescription use rates.
The National Ambulatory Medical Care Survey includes 2 e-prescribing questions: 1 question to assess computerized systems for prescription orders and a separate question to assess whether that system was used to send prescriptions electronically. The first question is, “Does your practice have a computerized system for orders for prescriptions?” Possible responses include “Yes,” “No,” “Don’t know,” or “Turned off.” Year 2007 was the first time that NAMCS asked this question in its surveys, and 2010 represents the most recent NAMCS data available. E-prescription use was assessed by a separate question that asked physicians, “If practice has a computerized system for orders for prescr iptions, are prescriptions sent electronically to the pharmacy?” Physicians could respond with “Yes,” “No,” “Unknown,” “Turned off,” or “Not applicable.” Responses other than “Yes” were considered as not having e-prescribing ability in this study. The original patient-level data were aggregated to physician-level data, which implies that each physician surveyed by the NAMCS was counted as only 1 data point.
Next, we appraised physician practice characteristics with respect to e-prescription access. The National Ambulatory Medical Care Survey allows physicians to categorize themselves into 1 of 14 medical specialties. Physicians were asked by NAMCS, “Is this a solo practice?”, for which we dichotomized the answers into “Yes” and “No.” Responses to the question “Type of Office Setting for This Visit” were divided into private practice, health maintenance organization (HMO), and other (community health center, mental health center, non–federal government clinic, and family planning clinic) to assess whether practice type was associated with varying rates of e-prescription use. The NAMCS design did not allow us to distinguish between small and large private practices. Physicians were also divided into owners or nonowners of their practices. We further analyzed differences in e-prescribing by geographical region (Northeast, South, Midwest, and West), consistent with the NAMCS methodology for grouping the geographic regions of ambulatory visits.
We used logistic regressions with the Heckman 2-step selection model to capture the 2-step question in NAMCS and consequent selection bias. The first step of the statistical model used a logistic regression to examine the factors that influenced whether a physician had a computerized system for prescription orders. With surveys that answered affi rmatively, a second logistic regression was used to examine factors that influenced whether prescriptions were sent electronically to the pharmacy.
Mill’s ratio was generated in the first logistic regression model. The inverse of this term was included as an extra independent variable in the second-step analysis to correct for the selection bias associated with e-prescription use. The inverse Mill’s ratio is essentially a transformation of the predicted individual probabilities in the first step, which can correct the selection bias by incorporating it as an additional independent variable. The second-step model illustrates Heckman’s insight that sample selection can be viewed as a form of omitted-variable bias, as conditional on both all original independent variables as well as on the inverse Mill’s ratio, as if the sample were randomly selected.9
In order to obtain national estimates, all analyses were weighted. In this study, the NAMCS sampling and weighting scheme has been omitted due to a large number of missing values, which resulted in the violation of simple random sampling assumption and would have led to incorrect standard errors. Alternatively, we adopted a weighting strategy that divided each weight by the average weight of the entire study sample.10 This weighting strategy—which allowed the NAMCS sampling weights to be incorporated without inflating the sample size—could approximate a simple random sampling and, in turn, correct variance estimation for logistic regressions. Data analysis was performed using Stata release 12 (StataCorp, College Station, Texas).
PDF is available on the last page.