Principal Findings
The opioid epidemic is a major concern in the United States, specifically in the Commonwealth of Pennsylvania [1,18] where the governor has issued a state emergency disaster declaration since 2018 [37]. Our hospitals and clinics are located in the heart of a major urban area within the state, in a county that hosts the highest estimated frequency of overdose deaths [38] and neighbors a district considered to be the epicenter for illicit opioid trade in the region [39].
In order to aid in many state-, city-, and hospital-level initiatives, our clinical informatics group has recently employed solutions to combat the opioid epidemic. These include simplifying PDMP queries, modifications of prescription settings in the emergency department, and analytics tools to track opioid prescribing throughout the institution [40,41].
Reduction in opioid prescriptions is a key step in combating this epidemic, and the duration of prescriptions has a significant effect on a patient’s risk of becoming a chronic opioid user [12,22].
In order to study our upcoming planned interventions, we realized we needed a reliable way to calculate durations of opioid prescriptions. Given the prior lack of requirement to document a written duration in our EHR system, we wanted to explore whether this was an accurate representation of a prescription’s true duration and whether making this a requirement improved overall representation of the duration of a prescription as measured by the calculated duration.
Overall, written duration was equal to the calculated duration (41.97%) or less than the calculated duration (41.06%) the majority of the time. However, 16.96% of the time our prescriptions had a longer calculated duration than the written duration, which contradicts CDC prescribing guidelines. We are concerned that unintentionally prolonged opioid prescription durations could be contributing to an increased risk of opioid dependence. We must take this opportunity to stress that we do not imply that these prescriptions were intentionally written to break guidelines or to prolong a patient’s duration of opioid therapy. We are simply stating that we observed a difference in these two values and we believe this requires attention. One solution to this issue could be modifications to the EHR system.
In order to provide peak levels of patient safety and end-user satisfaction, EHRs require continuous review and optimization. A lack of a required field allowed prescription durations to not match what the provider had documented, or likely intended. This is apparent in that requiring documentation of this field reduced those prescriptions without written durations, improved the percentage of written durations that accurately represented their corresponding calculated durations, and reduced the number of calculated durations that exceeded written durations. As discussed in our results, we explored the remaining 9.45% of prescriptions that continued to lack a written duration after our intervention and determined that they were likely due to renewals of prescriptions written prior to our modifications, where the system allows the field previously left blank to remain. Additionally, our intervention did yield a small but significant reduction in prescriptions where we were unable to compute a calculated duration (6.01% vs 3.61%). We attribute this to the addition of a quick button, as described in our methods.
Many studies have demonstrated how modifications to CPOE can affect opioid prescribing, though most of the literature focuses on tablet counts and not duration of treatment. Delgado et al demonstrated that when transitioning from an EHR that had no preset dispense quantity to an EHR that required a preset of 10 tablets, the median number of oxycodone 5 mg/ acetaminophen 325 mg tablets dispensed by two emergency departments decreased from 11.3 to 10.0 and from 12.6 to 10.9, respectively [33]. Similarly, Chiu et al demonstrated that when opioids were prescribed at discharge for outpatient surgery, modification of the default pill count from 30 tablets to 12 tablets reduced the median number of tablets dispensed from 30 to 20 tablets [34]. However, they did not examine prescription duration, noting that duration guidelines are usually “far longer than most patients will need,” and modifications of number of tablets dispensed will have a more profound affect [34]. While we believe reduction of tablets dispensed is critical to combating the opioid epidemic, at this time, CDC guidelines consider prescription duration to be an important measurement in reducing opioid therapy and combating the epidemic [23,24].
Conversely, other studies have demonstrated results from entirely removing presets for opioid tablets dispensed. Santistevan et al showed that in an emergency department setting, removal of the default of 20 tablets for hydrocodone and oxycodone, and the requirement of the prescriber to enter a number of tablets to dispense, reduced the median number of tablets prescribed from 20 to 15 [35]. The authors concluded that EHR presets may hinder providers’ ability to prescribe opioids to appropriately fit the variability in patient care (ie, more painful clinical conditions need more tablets) and that prescriptions written at the provider’s discretion may be more appropriate. Contrary to these results, Zwank et al demonstrated an increase in mean number of tablets from 15.31 to 15.77 after removing their 15-tablet preset and requiring manual entry of tablets dispensed [36].
Crothers et al also examined the effects of transitioning between EHR systems, from a homegrown system to a vendor-based system [42]. Their prior EHR system auto-calculated the maximum number of dispense units for PRN opioid prescriptions based on dose, frequency, and duration documented. During their implementation, they removed this functionality and instead developed preset dispense quantities, with outpatient prescriptions for clinics and inpatient discharges defaulted to 30 tablets for oxycodone and null for hydrocodone. This resulted in a decrease of 1.4 dispense units overall and 3.9 dispense units for inpatient discharge [42].
While these studies have examined the effect of changes in the requirement to document the number of tablets dispensed [33-36], to our knowledge we are the first to examine the requirement of documenting a duration field and comparing the duration intended by a prescriber (ie, written duration) to the calculated duration based on the number of tablets dispensed and instructions provided.
Our results confirm that simple documentation of an opioid prescription’s duration is not sufficient. When a duration field is not directly linked to other elements of the prescription and there is no automated calculation of duration from dose, frequency, and quantity—or vice versa—inaccuracies remain. While it is interesting that Crothers et al demonstrated that removal of such functionality reduced opioid dispensing [42], this was performed during a transition of EHR systems and with the addition of prescription presets, which likely influenced their results. Given the importance of the duration of an opioid prescription in the associated risk of long-term use, as well as further potential for overdose, it is vital to have data that accurately represents this value. Additionally, we have demonstrated that retrospective computation of a calculated duration is a viable alternative when written durations are not available for the study of EHR prescribing data.
In order to improve compliance to CDC opioid prescribing guidelines, EHR system vendors should consider rapidly developing tools that, by default, accurately represent prescription durations and/or modify doses and quantities dispensed based on provider-entered durations. As described in the Introduction, this calculation already exists in our system, but no hard stop or passive updating of prescription fields exists. If a provider intends to prescribe one tablet, three times a day for 3 days, the system should automatically set the quantity at nine tablets; if the provider reduces the quantity to six tablets, the duration of the prescription should automatically be reduced to 2 days.
Limitations
Our study was performed at a single, urban academic institution and its associated ambulatory clinics; therefore, our results may not represent the majority of the country’s hospitals. However, given that our institution is located in the state with the third-highest rate of drug overdose, our results may be valid when compared to other states of similar rates of overdose.
Additionally, as discussed in our methods, we intentionally biased our process of computing the calculated duration to make our resulting durations as short as possible by assuming individual doses were the higher of the possible range (ie, if a possible dose was 5-10 mg, we assumed the dose was 10 mg, thereby meaning that the quantity was used faster and the duration was shorter). This means that our calculations may underrepresent the number of prescriptions where the calculated duration was longer than the written duration if patients were to ration their medications and take lower doses over a longer period of time. Further, while our main intervention was making the duration field a requirement, we also added quick-action buttons and modified some prescription settings, which may have influenced our postperiod results, though we expect this to have been minimal. Finally, we used EHR data to represent durations of medications. While we believe it is important for our prescriptions to accurately represent the intended therapy and to abide by CDC guidelines, we did not determine whether individual patients took their prescribed medication as written, nor did we examine whether each prescription was filled at a pharmacy.