Abstract
Objective:
The present study assessed concordance in perioperative opioid fulfillment data between Michigan’s prescription drug monitoring program (PDMP) and a national pharmacy prescription database.
Summary Background Data:
PDMPs and pharmacy dispensation databases are widely utilized, yet no research has compared their opioid fulfilment data postoperatively.
Methods:
This retrospective study included participants (N=19,823) from two registry studies at Michigan Medicine between 7/1/2016 and 2/7/2019. We assessed concordance of opioid prescription fulfilment between the Michigan PDMP and a national pharmacy prescription database (Surescripts). The primary outcome was concordance of opioid fill data in the 91–180 day after surgical discharge, a time period frequently used to define persistent opioid use. Secondary outcomes included concordance of opioid dose and number of prescriptions fulfilled. Multinomial logistic regression analysis examined concordance across key sub-groups.
Results:
In total, 3,076 participants had ≥1 opioid fulfillments 91–180 days after discharge, with 1489 (49%) documented in PDMP only, 243 (8%) in Surescripts only, and 1,332 (43%) in both databases. Among participants with fulfillments in both databases, there were differences in the number (n=239; 18%) and dose (n=227; 17%). The PDMP database was more likely to capture fulfillment among younger and publicly insured participants, while Surescripts was more likely to capture fulfillment from counties bordering neighboring states. The prevalence of persistent opioid use was 10.7% using PDMP data, 5.5% using Surescripts data only, and 11.7% using both data resources.
Conclusions:
The state PDMP appears reliable for detecting opioid fulfillment after surgery, detecting two-times more patients with persistent opioid use compared with Surescripts.
Mini Abstract
This retrospective study compared two pharmacy dispensation databases and found a state-level prescription drug monitoring program identified two-times more patients with persistent opioid use following surgery compared to Surescripts, a national e-prescribing company.
Introduction
Surgery is a primary access point for opioid-based pain medications.1–3 Persistent opioid use (POU) has been identified as one of the most common complications after surgery with patients continuing to use opioids for 3–6 months after surgery.4–9 In recent years, vast changes to opioid prescribing policies took place at the national, regional, and institutional levels to prevent POU and other opioid-related problems.10–13 For example, multiple states have enacted policies requiring review of patient’s prescription drug monitoring program (PDMP) records prior to prescribing opioid analgesics.14 National private pharmacy databases have been integrated into the electronic health records and are also used for epidemiological studies and health services research. However, we do not know how accurately these data resources capture opioid use in the context of perioperative care.
Most research characterizing perioperative opioid use has relied on electronic health record data, limited to a single health care system, or insurance claims data, which lacks real-time updates or data for multiple payer types, as well as uninsured patients.15–17 Detecting POU is challenging clinically, given individuals fill beyond the usual window of postoperative care and may receive prescriptions from both inside and outside a single health system. As such, clinicians must rely on PDMPs to uncover more accurate fill patterns in real time across large geographic areas to inform clinical decision making.18,19 Increasingly, PDMPs and commercially-available pharmaceutical dispensation (i.e. Surescripts) databases have made their data available for research and quality improvement creating an opportunity to evaluate how well PDMPs perform clinically in detecting postoperative opioid use and characterizing exemplar clinical outcomes, such as POU for the first time.
The present study examined concordance in perioperative opioid prescription fulfillment data between two large pharmaceutical databases; Michigan’s PDMP and Surescripts, a national commercial data repository and pharmaceutical dispensation company. Surescripts manages nearly all electronic prescriptions (e-prescribing) in the Unites States by linking and transmitting prescriptions between electronic health records and pharmacies.20 We evaluated concordance between Michigan PDMP and Surescripts data for opioid fulfillment across pre-specified preoperative, perioperative, and postoperative time periods by linking these datasets with participant data in two registry studies of surgical patients in the state of Michigan. We then examined discrepancies in opioid fulfillment data at the level of patient and neighborhood characteristics to assess representation of patient opioid fulfillment in key subgroups. We also explored secondary outcomes including concordance of opioid dose, and number of opioid fulfillments.
METHODS
Data Sources and Patient Cohort
This retrospective, secondary data analysis study created the patient cohort originated from two prospective observational registries, the Analgesics Outcome Study (AOS)21–25 and Michigan Genomics Initiative (MGI; https://www.michigangenomics.org) at the University of Michigan. Data on prescription fulfillment for patients in the cohort were obtained from two pharmacy dispensation databases: Michigan’s PDMP and Surescripts.
AOS, started in 2010, is an on-going single-institution prospective observational cohort study of acute and chronic pain after elective total hip arthroplasty, total knee arthroplasty, hysterectomy, thoracic surgery, abdominal surgery, ankle surgery, breast surgery, hand surgery and inguinal hernia surgery. Started in 2012, MGI is an institutional repository of genetic and phenotypic data at the University of Michigan. In both AOS and MGI, patients aged 18 years of age older and scheduled to have elective surgery are recruited from the preoperative assessment clinic before surgery or in preoperative waiting area on the day of surgery. Patients who did not speak English, were unable to provide written informed consent and or were currently incarcerated are excluded from recruitment into these registries. Patients provided informed consent prior to participation and were informed that the study would access their electronic health record information and data could be used for future research, including linking to external sources of health information, such as prescription databases.
Michigan’s Automated Prescription Database, the PDMP, for the state of Michigan, has a record of all prescriptions that are filled for controlled substances (schedules 2–5) in Michigan and is used by the state to track fulfillment.26 Surescripts, connects prescribers with pharmacies, payers and pharmacy benefits managers across the US and is a commercial data repository of electronic prescription fulfillment dispensation. Surescripts reports it tracks virtually all electronic-prescriptions in the nation including 84% of all prescriptions in the US in 2020.20 Patient data obtained from AOS and MGI was linked to prescription PDMP and Surescripts databases. Data sharing agreements were obtained with State of Michigan and the PDMP vendor (Appriss Health) for the use of the PDMP data. This linkage of individual patients to their prescription fulfillment data was accomplished by an independent third-party data broker. The use of the data broker ensures that the study team only has access to an encrypted and de-identified patient and prescription dataset and the PDMP vendor only has access to the patient variables necessary for matching the patient to their prescriptions.
Patients were included in this study if they were enrolled in AOS or MGI and were had a surgical encounter between July 1, 2016, and February 7, 2019. Patients were excluded if they did not have a primary address in Michigan, as Michigan PDMP does not capture prescription fulfillment outside the state. Patients were excluded if they had subsequent surgery within the 180 days after discharge from their index surgery. Lastly, patients were also excluded if they had more than one match to a patient in PDMP data, per the independent third-party data broker. Only opioid prescriptions from PDMP and Surescripts were considered for analysis.
This study was determined to be exempt from regulation and the need for informed consent by the Institutional Review Board of the University of Michigan since it was a secondary analysis of deidentified data.
Outcomes and Explanatory Variables
The primary outcome of this study was a prescription fulfillment between 91–180 days after surgical discharge. This time period is previously defined for POU,27–29 and use of PDMP and/or commercial claims data would be more useful at delayed time postoperative points when the electronic health record is more likely to miss prescriptions. Secondary outcomes include other time periods important for defining POU and opioid-naïve status30,31: 1) preoperative period, defined as 180 days prior to admit for surgery to 31 days prior to admit for surgery 2) perioperative period, defined as 30 days prior to admit for surgery to 3 days post-discharge 3) an early postoperative period defined as 4–90 days post-discharge. Other secondary outcomes included number of opioid prescriptions fulfilled and total dose standardized to milligrams of oral morphine equivalents (OME).
Demographic data, including age, gender, race/ethnicity and were obtained from the electronic health record. Data on patient zip code to determine if the patient resided in a county that bordered another state, insurance (categorized as Blue Cross Blue Shield, other commercial, Medicare/Medicaid, and other which includes military, self-pay, workers comp, and unknown), inpatient surgery (categorized as length of stay >1 day ), comorbidities for the calculation of Charlson comorbidity index score, discharge date to determine time-period of discharge, and primary anesthesia Current Procedural Terminology (CPT) code to determine surgical body area were obtained from the electronic health record (EHR). A table of surgeries organized by surgical body area and the CPT codes used to determine surgical body area are in supplemental table 1 and 2, respectively.20 Missing data for race/ethnicity, patient insurance and primary anesthesia code was coded as Unknown.
Neighborhood data were extracted from the National Neighborhood Data Archive (NaNDA).32 NaNDA is a publicly available data archive that contains measures of socioeconomic and demographic characteristics by United States census tract. For this study, we used NaNDA-generated measures of ‘Neighborhood socioeconomic disadvantage,’ ‘Neighborhood affluence,’ and ‘Neighborhood ethnic immigrant concentration’ mapped to patient address.33
Statistical Analysis
Descriptive statistics were calculated for each patient characteristic based on whether the patient had an opioid prescription fulfilled in the PDMP only, in Surescripts only, in both the PDMP and Surescripts or neither database during the 91–180 days after surgical discharge. Descriptive statistics were also computed for number of opioid prescriptions fulfilled and opioid dose for patients who had a prescription fill in the PDMP only, in Surescripts only, in both PDMP and Surescripts during each of the 91–180 days after surgical discharge. Among patients who had a prescription fill in either or both PDMP and Surescripts, we calculated number and percentage of patients who had filled in PDMP only, in Surescripts only, and in both PDMP and Surescripts during the 91–180 post-discharge. Difference in proportions were evaluated using X2 difference test. For patients who fulfilled opioid prescriptions in both PDMP and Surescripts we calculated the discordance in the number of prescriptions and dose of prescriptions. We determined the dose to be discordant only if the lower OME was not within 90% of greater OME. A two-level multinomial logistic regression model with separate but correlated random effects was used to examine the factors associated with the likelihood of patients having a fill in the PDMP only, Surescripts only or both Surescripts and PDMP. Patients who did not fill a prescription during any of four primary or secondary time periods or did not have neighborhood data were not included in the model. POU was calculated using previously published definitions.34 All analyses were conducted using Stata version 15.1 (StataCorp) and 2-sided P < 0.05 was considered statistically significant.
RESULTS
Participant and Data Characteristics
A total of 24,207 participants enrolled in AOS or MGI had a surgery encounter between July 1, 2016, and February 7, 2019, and had data available through the PDMP and Surescripts databases. After excluding participants residing outside of Michigan, having another surgery in the 180 days after index surgery, and having more than one unique match in PDMP data, the final study cohort included 19,823 patients (see Figure 1). Descriptive data are included in Table 1.
Figure 1.
Study Cohort Diagram
Table 1.
Study cohort characteristics across pharmacy dispensation databases in late postoperative period (91 – 180 days post-surgical discharge).
| Overall | PDMP only N=1,489 | Surescripts only N=243 | PDMP and Surescripts N=1,332 | Neither dataset N=16,759 | |
|---|---|---|---|---|---|
|
|
|||||
| n (%) | n (%) | n (%) | n (%) | n (%) | |
|
| |||||
| Age in years, Mean (SD) | 53 (16) | 57 (14) | 53 (14) | 55 (14) | 53 (17) |
| Female | 10,816 (55) | 804 (54) | 142 (58) | 771 (58) | 9,099 (54) |
| Race/Ethnicitya | |||||
| White | 16,948 (85) | 1,312 (88) | 208 (86) | 1,157 (87) | 14,271 (85) |
| Black | 950 (5) | 74 (5) | 18 (7) | 85 (6) | 773 (5) |
| Asian | 319 (2) | 4 (0) | 0 (0) | 9 (1) | 306 (2) |
| Native Hawaiian and Other Pacific Islander | 16 (0) | 3 (0) | 0 (0) | 1 (0) | 12 (0) |
| American Indian/Alaskan native | 74 (0) | 4 (0) | 3 (1) | 7 (1) | 60 (0) |
| Other | 138 (1) | 4 (0) | 2 (1) | 6 (0) | 126 (1) |
| Hispanic | 398 (2) | 23 (2) | 2 (1) | 20 (2) | 353 (2) |
| Unknown | 980 (5) | 65 (4) | 10 (4) | 47 (4) | 858 (5) |
| Patient insurance | |||||
| Blue Cross Blue Shield | 10,100 (51) | 556 (37) | 98 (40) | 511 (38) | 8,935 (53) |
| Other commercial | 5,111 (26) | 437 (29) | 72 (30) | 418 (31) | 4,184 (25) |
| Medicare/Medicaid | 4,031 (20) | 423 (28) | 68 (28) | 371 (28) | 3,169 (19) |
| Other | 581 (3) | 73 (5) | 5 (2) | 32 (2) | 471 (3) |
| Border county | 1,811 (9) | 172 (12) | 36 (15) | 141 (11) | 1,462 (9) |
| Inpatient Surgery | 4,517 (23) | 497 (33) | 77 (32) | 410 (31) | 3,533 (21) |
| Outpatient Surgery | 15306 (77) | 992 (67) | 166 (68) | 922 (69) | 13,226 (79) |
| Charlson comorbidity score, Mean (SD) | 1.28 (2.34) | 2 (3) | 1 (2) | 2 (3) | 1 (2) |
| Time period of service | |||||
| Jul 16 - Dec 16 | 4,269 (22) | 379 (25) | 72 (30) | 335 (25) | 3,483 (21) |
| Jan 17 - June 17 | 3,903 (20) | 332 (22) | 55 (23) | 297 (22) | 3,219 (19) |
| Jul 17 - Dec 17 | 3,723 (19) | 272 (18) | 40 (16) | 234 (18) | 3,177 (19) |
| Jan 18 - Jun 18 | 4,370 (22) | 286 (19) | 54 (22) | 246 (18) | 3,784 (23) |
| Jul 18 - Feb 19 | 3,558 (18) | 220 (15) | 22 (9) | 220 (17) | 3,096 (18) |
| Neighborhood characteristics for patient primary addressb | |||||
| Neighborhood disadvantage, Mean (SD) | 0.09 (0.06) | 0.10 (0.07) | 0.11 (0.08) | 0.10 (0.07) | 0.08 (0.06) |
| Neighborhood affluence, Mean (SD) | 0.42 (0.18) | 0.35 (0.15) | 0.34 (0.15) | 0.37 (0.17) | 0.44 (0.18) |
| Ethnic and immigrant concentration, Mean (SD) | 0.05 (0.05) | 0.04 (0.04) | 0.04 (0.04) | 0.05 (0.04) | 0.05 (0.05) |
Note. PDMP = Prescription Drug Monitoring Program. SD = Standard Deviation.
All individuals of Hispanic ethnicity were included in Hispanic class. Other classes include non-Hispanic individuals only.
1640 patients did not have neighborhood characteristics available
Across both PDMP and Surescripts databases, a total of 5,052 had at least one opioid fill during the 91–180 after discharge. For secondary time periods, 5,365 patients had at least one opioid fill during the preoperative period, 13,880 had at least one opioid fill during the perioperative period, and 3,076 had at least one opioid fill during the early post-operative period.
Concordance of Opioid fill data in PDMP and Surescripts databases
Among individuals (N = 3,076) with any opioid fulfillment in the late postoperative period, between 91 to 180 after discharge, 1489 (49%) had ≥ 1 opioid prescriptions fulfilled in PDMP only, 243 (8%) had ≥ 1 fill in Surescripts only, and 1,332 (43%) had ≥ 1 documented in both databases. This pattern was consistent across all time periods (See Figure 2). Data for additional time periods shown in supplemental figure 1. Across databases there were differences in the number of opioid prescriptions fulfilled in the late postoperative period as well as the prescription size (Table 2). Among individuals with ≥ 1 opioid prescriptions fulfilled in both databases, 239 (18%) had discordant data related to number of prescriptions fulfilled, 227 (17%) had discordant data related to dose.
Figure 2.
Concordance of opioid fulfillment data in late postoperative period (91 – 180 days post-surgical discharge) in PDMP and Surescripts databases
Table 2.
Number of opioid prescriptions fulfilled and prescription size across databases during the late postoperative period (91 – 180 days post-surgical discharge)
| PDMP only N=1489 | Surescripts only N=243 | Surescripts and PDMP N=1332 |
||
|---|---|---|---|---|
| Surescripts | PDMP | |||
|
| ||||
| N (%) | N (%) | N (%) | N (%) | |
| Number of opioid prescriptions fulfilled | ||||
| 1 | 650 (44) | 76 (31) | 615 (46) | 522 (39) |
| 2 | 247 (17) | 37 (15) | 251 (19) | 244 (18) |
| 3 | 312 (21) | 83 (34) | 283 (21) | 304 (23) |
| 4 | 112 (8) | 21 (9) | 86 (6) | 106 (8) |
| 5 | 42 (3) | 5 (2) | 22 (2) | 43 (3) |
| 6 | 63 (4) | 12 (5) | 42 (3) | 54 (4) |
| 7 | 24 (2) | 3 (1) | 14 (1) | 25 (2) |
| 8 | 9 (1) | 2 (1) | 9 (1) | 15 (1) |
| 9 | 12 (1) | 1 (0) | 6 (0) | 8 (1) |
| ≥10 | 18 (1) | 3 (1) | 4 (0) | 11 (1) |
| Prescription size (OME), mean (SD) | 2436 (5157) | 4312 (8632) | 2188 (5169) | 2651 (7638) |
Note. OME = oral morphine equivalents; SD = Standard Deviation
Multinomial model
Table 3 presents the results of the two-level multinomial logistic regression. Having an opioid fill documented in both PDMP and Surescripts databases served as the reference group for the dependent variable. Results showed that opioid prescriptions filled during the perioperative period (relative to preoperative period) were more likely to be in PDMP only (RRR 3.29 [95% CI 2.87–3.74]), and less likely to be in Surescripts only (RRR 0.32 [95% CI 0.17–0.59]). Dataset concordance improved over time; relative to opioid fulfillment in 2016, opioid fulfillment was more likely to be represented in both databases (relative to PDMP or Surescripts only) in future time periods ranging from 2017–2019.
Table 3.
Results of multinomial logistic regression model (reference groups: opioid fulfillments in both PDMP and Surescripts datasets).
| Variable Categories | PDMP Only | Surescripts Only | ||||||
|---|---|---|---|---|---|---|---|---|
|
| ||||||||
| RRR | P value | 95% CI | RRR | P value | 95% CI | |||
|
| ||||||||
| Perioperative time period | ||||||||
| Preoperative | 1.00 | -- | -- | -- | 1.00 | -- | -- | -- |
| Perioperative | 3.28 | <0.01 | 2.87 | 3.74 | 0.32 | <0.01 | 0.17 | 0.59 |
| Early Postoperative | 0.88 | 0.09 | 0.76 | 1.02 | 0.65 | 0.18 | 0.34 | 1.23 |
| Late Postoperative | 1.13 | 0.19 | 0.95 | 1.34 | 0.93 | 0.85 | 0.46 | 1.88 |
| Age | 0.99 | 0.01 | 0.99 | 0.98 | 0.16 | 0.96 | 1.01 | 0.98 |
| Female | 1.15 | 0.10 | 0.97 | 1.14 | 0.71 | 0.58 | 2.25 | 1.14 |
| Race/Ethnicity | 1.00 | |||||||
| White | 1.00 | -- | -- | 1.00 | -- | -- | -- | |
| Black | 0.58 | <0.01 | 0.40 | 0.83 | 1.55 | 0.51 | 0.42 | 5.66 |
| Hispanic | 0.84 | 0.57 | 0.47 | 1.51 | 0.22 | 0.21 | 0.02 | 2.29 |
| Other | 1.89 | 0.04 | 1.03 | 3.46 | 14.73 | <0.01 | 3.45 | 62.93 |
| Unknown | 1.08 | 0.69 | 0.74 | 1.56 | 0.22 | 0.06 | 0.05 | 1.04 |
| Patient Insurance | ||||||||
| Blue Cross Blue Shield | 1.00 | -- | -- | -- | 1.00 | -- | -- | -- |
| Other commercial | 1.21 | 0.06 | 0.99 | 1.49 | 0.71 | 0.42 | 0.31 | 1.62 |
| Medicare/Medicaid | 1.34 | 0.02 | 1.04 | 1.74 | 1.02 | 0.97 | 0.41 | 2.57 |
| Other | 8.46 | <0.01 | 4.37 | 16.39 | 0.81 | 0.81 | 0.15 | 4.42 |
| Border County | 1.12 | 0.52 | 0.80 | 1.56 | 3.89 | <0.01 | 1.70 | 8.88 |
| Inpatient | 0.58 | <0.01 | 0.47 | 0.72 | 1.48 | 0.29 | 0.72 | 3.03 |
| Charlson index score | 1.06 | <0.01 | 1.02 | 1.10 | 1.02 | 0.76 | 0.89 | 1.17 |
| Period of Service | ||||||||
| Jul 16 – Dec 16 | 1.00 | -- | -- | -- | 1.00 | -- | -- | -- |
| Jan 17 - Jun 17 | 0.57 | <0.01 | 0.44 | 0.74 | 0.19 | <0.01 | 0.08 | 0.46 |
| Jul 17 - Dec 17 | 0.85 | 0.24 | 0.65 | 1.10 | 0.07 | <0.01 | 0.02 | 0.25 |
| Jan 18 - Jun 18 | 0.80 | 0.10 | 0.62 | 1.03 | 0.21 | <0.01 | 0.08 | 0.58 |
| Jul 18 - Feb 19 | 0.61 | <0.01 | 0.47 | 0.80 | 0.02 | <0.01 | <0.01 | 0.10 |
| Neighborhood Affluence | 0.18 | <0.01 | 0.09 | 0.38 | 0.02 | 0.03 | <0.01 | 0.64 |
| Neighborhood Disadvantage | 0.21 | 0.13 | 0.03 | 1.58 | 1.26 | 0.94 | <0.01 | 685.89 |
| Neighborhood Ethnic and Immigrant Concentration | 0.02 | <0.01 | <0.01 | 0.15 | 0.03 | 0.39 | <0.01 | 91.48 |
Note. CI = Confidence Interval; PDMP = Prescription Drug Monitoring Program; RRR = Relative Risk Ratio
The following patient and neighborhood factors increased the odds of having an opioid fill in PDMP only relative to both databases: having an unclassified race/ethnicity, having public or ‘other’ insurance, and having a higher Charlson comorbidity index score. The following factors increased the odds of having an opioid fill in Surescripts only relative to both databases: having an unclassified race/ethnicity and living in a county that bordered a neighboring state.
Conversely, the following factors decreased the odds of being in PDMP only: Black, non-Hispanic race/ethnicity (relative to White, non-Hispanic), older age, and having a length of stay longer than one day. The following factors decreased the odds of being in the PDMP only and Surescripts only (i.e., high database concordance): living in a more affluent neighborhood and living in a neighborhood with a higher concentration of diverse ethnicities and individuals who immigrated.
Persistent opioid use
Using PDMP data only, the prevalence of POU was 10.7% (2113/19823), and using Surescripts data only the prevalence was 5.5% (1096/19823). Using both databases, the prevalence of POU was 11.7% (2319/19823).
DISCUSSION
Overall, this study found Michigan’s PDMP identified approximately two-times more patients who fulfilled opioid prescriptions proximal to surgery and twice the number of individuals with POU compared to Surescripts, a commercial pharmacy dispensation company. There were also differences in the number of opioid fulfillments identified per person, and total opioid dose across the two databases. PDMP was more likely to capture opioid fill data during the perioperative period, for younger patients, for those with public insurance, and for those with ‘other’ insurance such as military, self-pay, and workers compensation. The one group more likely to be in Surescripts only relative to both databases was individuals who lived in a county that borders another state. Broadly, we found other participant and neighborhood characteristics were relatively concordant across PDMP and Surescripts databases.
Overall discordance in opioid fill data between the PDMP and Surescripts datasets likely reflects differences in the underlying characteristics of these two data sources. As in other states, reporting of opioid fills to the PDMP is legally mandated. In Michigan, the Board of Pharmacy Administrative Rule 338.3162b states all pharmacies, dispensing practitioners, and veterinarians who dispense schedules 2–5 controlled substances are required to electronically report this prescription data to PDMP on a daily basis. The data required by Rule 338.3162b shall be forwarded to the department by the end of the next business day and shall include data for all controlled substances dispensed since the previous transmission or report..26,35 Surescripts, is commercially-owned company with no legal mandates for reporting. As an e-prescribing company it does not track written opioid prescriptions, which were allowed in Michigan during the study period. This source of discordance will change with time, as Michigan is scheduled to enact an opioid e-prescribing mandate, and many other states already have these mandates in place.35 Notably, opioid fulfillment during the perioperative time period were more likely to be in PDMP only, and not in Surescripts which may reflect prescription fulfillment at hospitals and clinic pharmacies upon surgical discharge, when prescriptions are routed internally rather than sent to outside pharmacies.
While PDMP captured more opioid fulfillment than Surescripts by a two-fold margin, there were incidences of opioid fulfillment only documented in Surescripts. If we assume all incidences of opioid fulfillment in both datasets are true positives, the implication of this finding is that the PDMP has a margin of error with clinical and research implications. When calculating a clinical outcome like POU, adding Surecripts data to PDMP data increased the prevalence of POU from 10.7% to 11.7%, or an additional 206 cases of POU among 2,319 patients that would otherwise be overlooked. Thus, combining multiple data sources may be the most accurate method for clinical review and research.
While our cohort was limited to those with a primary address in Michigan county, those living in border counties were more likely to have data in Surescripts only, likely reflecting prescription fulfillment completed across state lines. This is a limitation of research and quality improvement uses of state-specific PDMPs, as PDMP data are regulated at a state level and policies with respect to access to the data vary between states. This is distinct from clinical uses of PDMPs where prescribers and pharmacists can easily access data from most states and the clinical queries routinely screen for out-of-state prescriptions. Michigan is a relatively large state, with limited border states; however, for some states with a higher ratio of border counties, this limitation of PDMP data could have a large impact on data accuracy. In these cases, Surescripts or neighboring state PDMP data could identify missed cases and therefore enhance the PDMP data.
This study has several strengths including the use of state- and national-level datasets linked to a large cohort of individuals undergoing surgery. Limitations of this study include analysis of a patient registry data from a single health system. The underlying cohort was predominantly White and non-Hispanic which limits generalizability of our findings to under-represented groups.36 State PDMPs vary widely in terms of how they are managed, administered, and used;37 thus results of this analysis may not apply to other states. The data capture from Surescripts may vary by state and region. Lastly, while our analysis enabled us to compare these databases, there was no external reference standard for validation because these two databases are the most representative databases of this outcome available.
Conclusions
Taken together these data suggest that overall, the state PDMP in Michigan is a more comprehensive data resource than Surescripts for capturing opioid fill data proximal to surgical events, although it may have limited accuracy in counties bordering other states. Clinically, this should be remedied by using clinical queries that screen for out-of-state prescriptions. Using PDMP and Surescripts as research data sources for studying opioid fulfillment proximal to surgery would yield vastly different incidence rates, clinical conclusions, and implications, and careful consideration should be given before using Surescripts alone for opioid prescription fulfillment data. This study underscores that PDMP databases offer highly relevant, real-world data for clinical review and studying opioid fulfillment and POU proximal to surgery.
Supplementary Material
Acknowledgements:
The authors acknowledge the Michigan Genomics Initiative participants and the Precision Health Initiative at the University of Michigan.
Conflicts of Interest and Sources of Funding:
Dr Brummett is a consultant for Heron Therapeutics and Vertex Pharmaceuticals. He has performed 1-time advisory roles for Alosa Health and the Benter Foundation. In addition, he provides expert testimony for medical malpractice. Grant support for this project was provided by the National Institute of Drug Abuse R01DA042859 and Precision Health Initiative at the University of Michigan.
References
- 1.Liu S Nonfatal Drug and Polydrug Overdoses Treated in Emergency Departments — 29 States, 2018–2019. MMWR Morb Mortal Wkly Rep. 2020;69:1149–1155. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Wilson N Drug and Opioid-Involved Overdose Deaths — United States, 2017–2018. MMWR Morb Mortal Wkly Rep. 2020;69:290–297. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Hill M V, Mcmahon ML, Stucke RS, et al. Wide variation and excessive dosage of opioid prescriptions for common general surgical procedures. Ann Surg;265. Epub ahead of print 2017. DOI: 10.1097/SLA.0000000000001993. [DOI] [PubMed] [Google Scholar]
- 4.Brummett CM, Waljee JF, Goesling J, et al. New persistent opioid use after minor and major surgical procedures in us adults. JAMA Surg. 2017;152:170504. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Gossett TD, Finney FT, Hu HM, et al. New Persistent Opioid Use and Associated Risk Factors Following Treatment of Ankle Fractures. Foot ankle Int. 2019;40:1043–1051. [DOI] [PubMed] [Google Scholar]
- 6.Brescia AA, Waljee JF, Hu HM, et al. Impact of Prescribing on New Persistent Opioid Use After Cardiothoracic Surgery. In: Annals of Thoracic Surgery. Elsevier; USA; 2019:1107–1113. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Tam CA, Dauw CA, Ghani KR, et al. New Persistent Opioid Use After Outpatient Ureteroscopy for Upper Tract Stone Treatment. Urology. Epub ahead of print December 1, 2019. DOI: 10.1016/j.urology.2019.08.042. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Johnson SP, Chung KC, Zhong L, et al. Risk of Prolonged Opioid Use Among Opioid-Naïve Patients Following Common Hand Surgery Procedures. J Hand Surg Am. 2016;41:947–957.e3. [DOI] [PubMed] [Google Scholar]
- 9.Harbaugh CM, Lee JS, Hu HM, et al. Persistent Opioid Use Among Pediatric Patients After Surgery. Pediatrics. 2018;141:e20172439. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Vu J V, Howard RA, Gunaseelan V, et al. Statewide Implementation of Postoperative Opioid Prescribing Guidelines. https://doi.org/101056/NEJMc1905045. 2019;381:680–682. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Opioid Resources - Laws and Regulations Available from: https://www.michigan.gov/opioids/0,9238,7-377-88143_88345---,00.html. Accessed September 8, 2021.
- 12.Dowell D, Haegerich TM, Chou R. CDC guideline for prescribing opioids for chronic pain-United States, 2016. JAMA - J Am Med Assoc. 2016;315:1624–1645. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.MC B, J W, AC F. Unintended Consequences From the 2016 US Centers for Disease Control and Prevention Guideline for Prescribing Opioids-Accelerating Change in Postoperative Prescribing. JAMA Netw open;4. Epub ahead of print June 11, 2021. DOI: 10.1001/JAMANETWORKOPEN.2021.11997. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.The Pew Charitable Trusts. When Are Prescribers Required to Use Prescription Drug Monitoring Programs? Available from: https://www.pewtrusts.org/en/research-and-analysis/data-visualizations/2018/when-are-prescribers-required-to-use-prescription-drug-monitoring-programs. Accessed November 9, 2021.
- 15.Karhade A V, Schwab JH, Bedair HS. Development of Machine Learning Algorithms for Prediction of Sustained Postoperative Opioid Prescriptions After Total Hip Arthroplasty. J Arthroplasty. 2019;34:2272–2277.e1. [DOI] [PubMed] [Google Scholar]
- 16.Harbaugh CM, Nalliah RP, Hu HM, et al. Persistent opioid use after wisdom tooth extraction. JAMA - J Am Med Assoc. 2018;320:504–506. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Sekhri S, Arora NS, Cottrell H, et al. Probability of Opioid Prescription Refilling After Surgery: Does Initial Prescription Dose Matter? Ann Surg. 2018;268:271–276. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Perrone J, Nelson LS. Medication Reconciliation for Controlled Substances — An “Ideal” Prescription-Drug Monitoring Program. http://dx.doi.org.proxy.lib.umich.edu/101056/NEJMp1204493. 2012;366:2341–2343. [DOI] [PubMed] [Google Scholar]
- 19.AT M, C G, R L, et al. Characteristics of state prescription drug monitoring programs: a state-by-state survey. Pharmacoepidemiol Drug Saf. 2016;25:847–851. [DOI] [PubMed] [Google Scholar]
- 20.National Progress Report 2020. | Surescripts Available from: https://surescripts.com/news-center/national-progress-report-2020. Accessed November 9, 2021.
- 21.Janda AM, As-Sanie S, Rajala B, et al. Fibromyalgia Survey Criteria Are Associated with Increased Postoperative Opioid Consumption in Women Undergoing Hysterectomy. Anesthesiology. 2015;122:1103–1111. [DOI] [PubMed] [Google Scholar]
- 22.Brummett CM, Urquhart AG, Hassett AL, et al. Characteristics of fibromyalgia independently predict poorer long-term analgesic outcomes following total knee and hip arthroplasty. Arthritis Rheumatol. 2015;67:1386–1394. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Hilliard PE, Waljee J, Moser S, et al. Prevalence of Preoperative Opioid Use and Characteristics Associated With Opioid Use Among Patients Presenting for Surgery. JAMA Surg. 2018;153:929–937. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Goesling J, Moser SE, Zaidi B, et al. Trends and predictors of opioid use after total knee and total hip arthroplasty. Pain. 2016;157:1259–1265. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Brummett CM, Janda AM, Schueller CM, et al. Survey criteria for fibromyalgia independently predict increased postoperative opioid consumption after lower-extremity joint arthroplasty: a prospective, observational cohort study. Anesthesiology. 2013;119:1434–1443. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Michigan Licensing and Regulatory Affairs. Data Submitters Michigan Automated Prescription System Available from: https://www.michigan.gov/lara/bureau-list/bpl/health/maps/data-submitters. Accessed April 8, 2022.
- 27.Ladha KS, Patorno E, Liu J, et al. The Impact of Perioperative Epidural Placement on Post-Discharge Opioid Use in Patients Undergoing Abdominal Surgery. Anesthesiology. 2016;124:396. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Qureshi R, Werner B, Puvanesarajah V, et al. Factors Affecting Long-Term Postoperative Narcotic Use in Discectomy Patients. World Neurosurg. 2018;112:e640–e644. [DOI] [PubMed] [Google Scholar]
- 29.Cancienne JM, Patel KJ, Browne JA, et al. Narcotic Use and Total Knee Arthroplasty. J Arthroplasty. 2018;33:113–118. [DOI] [PubMed] [Google Scholar]
- 30.Howard R, Fry B, Gunaseelan V, et al. Association of Opioid Prescribing with Opioid Consumption after Surgery in Michigan. JAMA Surg. 2019;154:184234. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Jivraj NK, Raghavji F, Bethell J, et al. Persistent Postoperative Opioid Use: A Systematic Literature Search of Definitions and Population-based Cohort Study. Anesthesiology. 2020;132:1528. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Nanda (National Neighborhood Data Archive) | Available from: https://nanda.isr.umich.edu/. Accessed October 8, 2021.
- 33.Morenoff DJD, House JS, Hansen BB, et al. Understanding Social Disparities in Hypertension Prevalence, Awareness, Treatment, and Control: The Role of Neighborhood Context. Soc Sci Med. 2007;65:1853. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Howard R, Gunaseelan V, Brummett C, et al. New Persistent Opioid Use After Inguinal Hernia Repair. Ann Surg.;Publish Ah. Epub ahead of print 2020. DOI: 10.1097/sla.0000000000004560. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Michigan Legislature - Section 333.7333a Available from: http://www.legislature.mi.gov/(S(5v3pzgfvr2sy21swq1o4cowj))/mileg.aspx?page=getObject&objectName=mcl-333-7333a&highlight=7333a. Accessed October 8, 2021.
- 36.Spector-Bagdady K, Tang S, Jabbour S, et al. Respecting Autonomy And Enabling Diversity: The Effect Of Eligibility And Enrollment On Research Data Demographics. Health Aff (Millwood). 2021;40:1892–1899. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Moyo P, Simoni-Wastila L, Griffin BA, et al. Prescription drug monitoring programs: Assessing the association between “best practices” and opioid use in Medicare. Health Serv Res. 2019;54:1045. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.