The Impact of Electronic Health Record Use on Physician Productivity
Published Online: November 25, 2013
Julia Adler-Milstein, PhD; and Robert S. Huckman, PhD
The centerpiece of the 2009 Health Information Technology for Economic and Clinical Health (HITECH) Act is $27 billion in incentives for providers who demonstrate “meaningful use” of electronic health records (EHRs).1 The legislation was motivated by the belief that EHRs used in specific ways (eg, medication order entry with alerts for drug-drug interactions) would make care safer, more effective, and more efficient. The meaningful use measures specify the activities that must be performed using an EHR,2 leaving ambulatory practices and hospitals to determine how to structure their work to accomplish them.
A major concern among physicians is that EHR adoption will hamper their productivity.3 This concern is not without merit—several studies have shown that physicians spend extra time entering data into the EHR, which cuts into time with patients and can extend the length of the workday.4,5 One strategy for dealing with this productivity loss is to rely on support staff to perform EHR-related tasks. However, when work is interdependent, delegation has its own costs; it increases the need for coordination, which may require additional physician time.6 Because scope-of-practice regulations prevent support staff from performing many clinical activities autonomously, physicians who successfully delegate must still spend time reviewing and authorizing staff activities.
For example, a physician who uses support staff to enter orders into an EHR must review and sign them before they are submitted. It is unclear whether (or under what conditions) efficiency gains from delegation exceed the time required to communicate the orders that need to be entered, reviewed, and in some cases, corrected. Across all clinical activities, there is little empirical evidence to guide physicians about whether it is optimal to off-load EHR-related tasks to support staff, or whether doing so would make them less efficient due to costs of oversight and coordination.7
Our study used monthly EHR task-log data from more than 40 primary care practices to examine the relationship between physician productivity, the degree of EHR use, and the delegation of EHR tasks. We first examined the independent effects of EHR use and delegation on productivity, and then assessed their joint impact on productivity to shed light on whether delegation and EHR use operate as complements or substitutes. Finally, we explored whether these relationships differ by practice size. Our findings offer insight into how primary care practices can structure their work after adopting an EHR to ensure that EHR use does not harm productivity.
Sample, Data and Measures
We obtained panel data for all of the primary care and internal medicine practices (n = 42) that use both a web-based EHR and a billing and practice management system from athenahealth Inc (Watertown, Massachusetts). Practices were distributed throughout the country and had on average 4 clinicians (range, 1-14). The average length of time that practices used the EHR was 17 months, with a minimum of 6 months. All practices in our sample employed at least 1 clinical support staff member and therefore had the ability to delegate from clinicians (those who can bill for clinical services, predominantly physicians but also including nurse practitioners and physician assistants) to clinical support staff (eg, registered nurse, licensed practical nurse, medical assistant). Our data included monthly measures at the practice level from May 2006 through May 2009. Observations were included for each practice with at least 6 months of experience, beginning in the first full month after it adopted the EHR and ending in May 2009, when athenahealth created the data set (n = 695 practice-month observations).
We relied on data from 2 sources: (1) the billing and practice management software and (2) the EHR. The billing and practice management software captures practice and staff demographics, monthly appointment volume, and monthly billing data. The EHR tracks each of hundreds of discrete, time-stamped actions associated with patient care. These actions are best thought of as changes to fields within the EHR. For descriptive purposes, they are grouped into a more meaningful and manageable list of 32 clinical tasks. For example, the task “collect vitals” includes 30 fields (eg, height, weight) and a change to any field is captured as a distinct action (eAppendix Table 1, available at www.ajmc.com). The vendor then generated a data set with a count of the number of actions, grouped by task, performed by each staff member per month. We used these 2 data sources to create the measures of productivity, EHR use, and delegation described below.
Productivity. Physician efficiency and productivity have been defined and studied in a number of ways.8-16 In the context of EHRs, physician productivity is most commonly used to refer to throughput—the number of services delivered in a given period of time. Physicians carefully weigh whether to make practice changes (eg, EHR adoption) based on their perception of whether the change affects this dimension of productivity. We measured monthly productivity at the practice level using work relative value units (RVUs), standardized units of production in healthcare that reflect the volume and intensity of services provided, and serve as the basis for fee-for-service reimbursement. Work RVUs are captured in the billing and practice management system, and include all work RVUs for which the practice billed, regardless of whether they were reimbursed by the specific payer. For each practice, we divided total work RVUs per month by the number of clinician workdays in the month. We then log transformed this variable to approximate a normal distribution. (We ran all our models with the untransformed version of the dependent variable to confirm that no results were driven by the transformation.)
EHR Use. To capture the degree of EHR use, we created a measure of average task frequency per appointment by dividing (1) the total number of actions across all tasks in the EHR conducted by all staff members in each month by (2) the number of appointments in the month. Our measure of EHR use relied on data taken directly from the EHR, eliminating any potential self-reporting bias, and was not limited to specific EHR functions such as order entry.17-19 Our measure therefore captured a comprehensive picture of use. Our measure was also granular, capturing the distinct actions taken in the EHR, not simply whether the EHR was used. For example, the entry of each vital sign (eg, weight, height, blood pressure) was captured as a distinct action. That resulted in a robust measure of the average intensity of EHR use per visit in the month.
(As used in our models [described below], the measure relies on the identifying assumption that the underlying distribution of work per visit is consistent within a practice in a given calendar month. Therefore, increases in EHR use reflected greater use of the system to document work that was performed as opposed to reflecting increases in the volume of work itself. We empirically assessed the validity of this assumption by examining whether the number of monthly appointments predicted EHR use, using our base model specification. We found no evidence that more appointments were associated with higher levels of EHR use, suggesting that increased EHR use was not a reflection of increased work, but instead a reflection of increased documentation of work performed. However, because EHR use could also be a function of variation in the intensity with which patients are treated, we included month fixed effects to accommodate average changes in treatment intensity associated with seasons [eg, the flu season], and we included practice fixed effects under the assumption that patient treatment intensity is constant over time within practices.)
Delegation. Our delegation measure, calculated for each practice month and reported as a percentage, was the total number of actions across all tasks in the EHR conducted by clinical support staff divided by the total number of actions across all tasks in the EHR conducted by either clinical support staff or clinicians.
Interaction Between Delegation and EHR Use. To assess whether EHR use and delegation operate as complements or substitutes in their impact on productivity, we created a continuous interaction term by first centering each measure around its mean (to improve interpretability) and then multiplying it. A positive coefficient on this term reflects a complementary impact on productivity, and a negative coefficient reflects a substitutive joint impact on productivity.
Practice Size. To assess whether the relationships of interest varied by practice size, we split our sample in half (21 practices in each group), effectively splitting practices into those with 1 to 3 clinicians and those with 4 or more clinicians. In other analyses, we split our sample into thirds (14 practices in each group) with practice groups of 1 to 2 clinicians, 3 to 5 clinicians, and 6 or more clinicians. We interacted sizebased dichotomous variables with EHR use, delegation, and the continuous interaction term. (Small [1-3 clinicians] and large [4+ clinicans] practices were not statistically different in terms of the length of time that they had used the system. For small practices, the average was 17.36 months; for large practices, the average was 16.05 months [P = .60 for difference in means].)
Analytic Approach. We used an ordinary least squares model that estimated log work RVUs per clinician workday within each practice-month as a function of the intensity of EHR use and the level of delegation. We then added the interaction between EHR use and delegation to the model above. In a final model, we included a 3-way interaction, multiplying our categorical measures of practice size by EHR use, delegation, and the interaction term. The interaction terms assessed whether EHR use and delegation operate as complements or substitutes; that is, whether the productivity impact of increasing one element increases (complement) or decreases (substitute) the impact of the other. Standard errors were clustered at the practice level. (Please refer to the eAppendix Regression Model at www.ajmc.com for our regression equation.)
All models included fixed effects at the practice level to control for practice-specific, time-invariant factors that might affect productivity (eg, a practice that includes both physicians and nurse practitioners compared with a practice composed solely of physicians). Additional controls included an indicator for the first 6 months following EHR adoption to allow for an implementation period in which work patterns may be in flux; a measure of the proportion of clinicians in the practice under 35 years of age to account for the possibility that recent medical school graduates may be more skilled EHR users; and the number of clinical and administrative support staff per clinician in the practice—each of which may influence the level of delegation as well as productivity. (Since we were concerned that controlling for the number of support staff would negate the effect of a practice that chose to increase support staff in order to delegate more, we re-ran our models without these 2 variables. Our results did not materially differ and the results we present include them.)
We used the subset of tasks for which there is a “collect” (ie, newly entered) and a “review” (ie, viewed but not changed) option (eg, allergies; Table 1) to control for the monthly ratio of new to existing data in the EHR. This adjusted for the extent to which data need to be newly entered in the month, which could result from a changing patient mix or the approach of transitioning from paper to electronic records (eg, up-front conversion of all records, gradual conversion upon first patient visit), both of which may influence the intensity of EHR use and productivity. We included a full set of calendar month indicators to account for the possibility that productivity might be affected by seasonal changes in patient volume and illness severity (eg, flu season, holidays).
PDF is available on the last page.