CPOE includes computerized ordering of imaging tests, laboratory orders, referrals to other providers, and sometimes also electronic medication prescribing, though e-prescribing is often studied as a separate functionality. In Meaningful Use criteria [7], as well as in the survey data we use (see below), e-prescribing is also measured by separate items. We focus in this paper on the Healthcare Information and Management Systems Society (HIMSS) Analytics Ambulatory Survey (HIMSS Analytics LOGIC™ Market Intelligence Platform) questions that ask about CPOE and not the specific medication ordering questions, and we interpret the survey questions about CPOE to include laboratory and other tests and referrals that a doctor can order.
Data source
We used three-year panel data (2014–2016) from the annual HIMSS Analytics Ambulatory Survey. This is a companion survey to the original HIMSS hospital survey, which has been used in studies of HIT and CPOE adoption [8,9,10]. The ambulatory survey includes clinics defined as facilities that provide “preventative, diagnostic, therapeutic, surgical, and/or rehabilitative outpatient care where the duration of treatment is less than 24 hours—and is generally referred to as outpatient care.”
The survey captures information on more than 75% of U.S. health system-associated ambulatory care practices [11]. HIMSS defines a health system as an organization including at least one hospital and its associated nonacute facilities, where “associated” indicates a governance relationship (i.e., owned, leased, or managed by a health system). The survey includes approximately 42,000 clinics across the United States.
Approach
We limited our datasets to health system-associated ambulatory clinics that deliver primary or specialty care, eliminating imaging centers or other locations where ordering was not expected. We removed observations that were missing entries for predictor variables. When adoption patterns seemed illogical or inaccurate—for example, if a clinic was noted as using CPOE in 2014 and 2016 but not in 2015, or if an answer did not match the number of answer choices, we considered the responses as probably incorrect and dropped the clinic from our sample. This logic-checking resulted in eliminating 2169 records, or about 10%, with a final sample size of 19,109 clinics.
Clinic characteristics
Using HIMSS survey data, we classified ambulatory clinics according to size (based on number of physicians, dichotomized into 3 or fewer vs. > 3); clinic type (primary versus specialty), and health system type (single hospital or multi-hospital health system).
Study outcome
We measured two outcomes, CPOE use and the frequency of that use, that were collected in the HIMSS survey. The survey first asks clinics about how EHR software is being used, with a checkbox option for “Clinician Order Entry,” among other options. Within the Clinician Order Entry question response options, respondents could indicate location where CPOE was used (i.e., whether the functionality was available at a clinician station or at the point of care) with both options potentially being selected. We created a single variable that represented overall use, whether the overall question or any location was selected, and we used this variable for most of our analyses.
The second relevant question focused on percent of CPOE use, titled “CPOE - % of Medical Orders Entered by Physicians.” If this option was checked off in the main column, the respondent was asked to estimate a percentage of medical orders that were entered electronically by physicians. There were five options: 1–25% of orders, 26–50% of orders, 51–75% of orders, 76–94% of orders, or > 94% of orders. We used these findings for analyses about percentage of orders.
The survey did not clarify whether CPOE was to include e-prescribing, and there were other medication management questions on the survey, so we did an additional analysis on the questions on electronic medication prescribing (namely, two checkboxes for “E-Prescribing new medications” and “E- Prescribing refill medication requests”) as a comparison, described below.
Analysis
We computed descriptive statistics, reporting the mean CPOE use rate by year (2014–2016) and by clinic type.
We used a multivariable linear probability model to examine the association between use of CPOE by 2016 and three clinic characteristics available in the HIMSS database: size (based on number of physicians, dichotomized into 3 or fewer vs. > 3); clinic type (primary versus specialty), and whether the clinic was part of a health system, and if so what system type (i.e., single hospital or multi-hospital health system).
As a sensitivity analysis, we examined the factors associated with changes in adoption status. With the subsample of clinics that did not have CPOE in 2014, we used multivariable linear probability models to analyze the relationship between size, clinic type, and health system type and new adoption (i.e., clinics newly adopting CPOE in 2015 or 2016). In all regression analyses, we clustered standard errors at the health system level to account for multiple clinics associated with the same health system.
We then examined use of CPOE by clinics within health systems by calculating the number of health systems with full adoption (all affiliated clinics used CPOE), partial adoption (some clinics used CPOE while others did not), or no adoption (no clinics within a system used CPOE).
As additional analyses, we explored factors predicting use of electronic prescribing of medications, to see if there were different predictors for this related functionality, which is a form of CPOE restricted to medications, and we also compared data exchange over time and among those with CPOE and without.