Association of Primary Care Clinic Appointment Time With Opioid Prescribing

Key Points Question Is the decision to prescribe opioids associated with appointments that are behind schedule or later in the day compared with earlier or on-time appointments? Findings In this cross-sectional study, opioid prescribing for opioid-naive patients with pain diagnoses was significantly associated with increases as the workday progressed and with appointments that started late, although the effect size was modest. Nonopioid pain treatment orders did not show similar patterns. Meaning Appointment timing that contributes to time pressure could be adversely associated with physician decision-making and could have widespread relevance for public health and quality improvement efforts, if similar patterns exist in other clinical scenarios.


Study Sample Definition
To define our study sample, we employed appointment-level, patient-level, and physician-level inclusion criteria. At the appointment level, we limited our sample to the more than 95 percent of appointments scheduled for the four most common durations (10, 15, 20, or 30 minutes) on a weekday (Monday-Friday). Since we relied on time stamps recorded by the EHR during the visit to calculate appointment order and construct other variables, we also restricted our sample based on the quality of each appointments timing data. Specifically, we excluded the small number of appointments with no recorded start or stop time and appointments where the dates for the start and stop time stamps did not match. Finally, we excluded appointments with an encounter starting time stamp more than three hours different from the scheduled visit time, as this likely indicates non-real-time use of the EHR (i.e., delayed appointment documentation).
Our diagnosis-related appointment-level sample restrictions limited the sample to those visits where the provider recorded a diagnosis of pain during the appointment and the patient did not have a cancer diagnosis [140.x-239.x]. We defined pain using the CDC's classification of ICD-9 codes that indicate back or chronic pain. We grouped chronic pain diagnoses into the following categories listed in eTable 1. We limited our sample to appointments with the first occurrence of any diagnosis within a category (e.g., back pain) within the past year, using 2016 claims data.
At the patient level, we excluded pediatric appointments by restricting our sample to visits where the patient was at least 20 years of age. While this exceeds the legal definition of a minor, our data contained age information in 10-year increments. We also limited our analyses to © 2019 Neprash HT et al. JAMA Network Open.
patients who were opioid naïve at the time of appointmentdefined as having no recorded opioid prescription from an athenahealth provider during the past 365 days.
At the provider level, we included physicians (MDs and DOs) with a primary care specialty, including internal medicine, family practice, and general practice. We excluded a small number of physicians who did not prescribe any opioids during our study period, as this likely indicates that they do not use computerized physician order entry (CPOE).

Opioid Prescriptions
We identified opioid prescriptions as any CPOE item listed as a prescription with an enhanced therapeutic class indicating an analgesic narcotic. This included the generic ingredients listed in eTable 2. We excluded opioids designed to treat opioid use disorder or opioid dependence, identified as anything containing the following strings of text: 'methado', 'dolophine', 'buprenorphine', 'subutex', 'suboxone', 'buprenex', 'butrans', 'probuphine', or 'belbuca'. We also excluded the small fraction of prescriptions that were not denominated in milligrams or tablet quantities. We matched opioid prescriptions to appointments using patient identifier, physician identifier, and date.

Non-Opioid Prescriptions
We identified NSAIDs (see eTable 2 for generic ingredients) and statins by searching the enhanced therapeutic class (ETC) label for the '%nsaid%' and '%antihyperlipid%', respectively.

Physical Therapy Referals
From the universe of all non-prescription orders placed through the EHR, we identified physical therapy referrals as those where the clinical order type name contained the text string '%physical therapy%'.

Chronic Condition Indicators
We defined 27 chronic condition indicators according to the algorithm used by the Chronic Conditions Data Warehouse. When possible, we used the reference period specified in the algorithm. Athenahealth data likely does not include complete claims for all patients, so we anticipate that our measures under capture the presence of chronic conditions within our sample.
We rely not on individual indicators, but on a summary measure that counts how many conditions are present.

Regression Specifications
Our main analysis estimated the following linear regression specification: indexing appointment i with physician j at time t. Outcomes of interest were prescribing (opioid, NSAID, antihypertensives, statins) and referral to physical therapy. is appointment rank within each physician's workday. is a vector of patient and appointment characteristics including patient age category, sex, insurer, chronic condition count, pain type, and scheduled appointment duration. are physician-practice fixed effects (in reality, these are likely physician fixed effects, but we cannot observe if a physician practices within multiple parent organizations). are season-year (e.g., Jan-Mar, 2017) fixed effects. Analyses of appointment lateness used a similar equation with appointment start hour as an additional control variable. eFigure 2. Opioid Prescribing by Appointment Order, Odds Ratios NOTE: Point estimates and confidence intervals were estimated using multivariate logistic regression models with opioid prescribing as the dependent variable. The key covariate was an indicator for appointment order within the day. The model was adjusted as described in Figure 1, with physician random effects, rather than fixed effects. eFigure 3. Opioid Prescribing by Appointment Lateness, Odds Ratios NOTE: Point estimates and confidence intervals were estimated using multivariate logistic regression models with opioid prescribing as the dependent variable. The key covariate was an indicator for appointment lateness. The model was adjusted as described in Figure 1, with physician random effects, rather than fixed effects. eFigure 4. Antihypertensive and Statin Prescribing, by Appointment Order NOTE: Point estimates and confidence intervals were estimated with the use of multivariate linear regression models with antihypertensive and statin prescribing as the dependent variables.
The key covariate was an indicator for appointment order. Both models were adjusted as described in Figure 1. eFigure 5. Antihypertensive and Statin Prescribing, by Appointment Lateness NOTE: Point estimates and confidence intervals were estimated with the use of multivariate linear regression models with antihypertensive and statin prescribing as the dependent variables.
The key covariate was an indicator for appointment lateness category. Both models were adjusted as described in Figure 1. Note: Point estimates and standard errors were estimated with the use of multivariate linear regression models with opioid prescribing as the dependent variable. Column (2) adjusted for physician fixed effects. Column (3) was adjusted as described in Figure 1. *** p<0.01, ** p<0.05, * p<0.10