Cycles of Chronic Opioid Therapy Following Mandatory Prescription Drug Monitoring Program Legislation: A Retrospective Cohort Study

Abstract

Background

Mandates for prescriber use of prescription drug monitoring programs (PDMPs), databases tracking controlled substance prescriptions, are associated with reduced opioid analgesic (OA) prescribing but may contribute to care discontinuity and chronic opioid therapy (COT) cycling, or multiple initiations and terminations.

Objective

To estimate risks of COT cycling in New York City (NYC) due to the New York State (NYS) PDMP mandate, compared to risks in neighboring New Jersey (NJ) counties.

Design

We estimated cycling risk using Prentice, Williams, and Peterson gap-time models adjusted for age, sex, OA dose, payment type, and county population density, using a life-table difference-in-differences design. Failure time was duration between cycles. In a subgroup analysis, we estimated risk among patients receiving high-dose prescriptions. Sensitivity analyses tested robustness to cycle volume considering only first cycles using Cox proportional hazard models.

Participants

The cohort included 7604 patients dispensed 12,695 prescriptions.

Interventions

The exposure was the August 2013 enactment of the NYS PDMP prescriber use mandate.

Main Measures

We used monthly, patient-level data on OA prescriptions dispensed in NYC and NJ between August 2011 and July 2015. We defined COT as three sequential months of prescriptions, permitting 1-month gaps. We defined recurrence as re-initiation of COT after at least 2 months without prescriptions. The exposure was enactment of the PDMP mandate in NYC; NJ was unexposed.

Key Results

Enactment of the NYS PDMP mandate was associated with an adjusted hazard ratio (HR) for cycling of 1.01 (95% CI, 0.94–1.08) in NYC. For high-dose prescriptions, the risk was 1.16 (95% CI, 1.01–1.34). Sensitivity analyses estimated an overall risk of 1.01 (95% CI, 0.94–1.11) and high-dose risk of 1.09 (95% CI, 0.91–1.31).

Conclusions

The PDMP mandate had no overall effect on COT cycling in NYC but increased cycling risk among patients receiving high-dose opioid prescriptions by 16%, highlighting care discontinuity.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11606-022-07551-z.

INTRODUCTION

Non-medical opioid analgesic (OA) use constitutes an epidemic in the United States (US). In 2019, over 9.7 million individuals in the US aged 12 and older reported past-year non-medical OA use.¹ Increases in non-medical OA use were accompanied by increases in OA-involved overdose deaths. Over 14,000 overdose deaths in the US involved OAs in 2019, constituting over 28% of overdose deaths.²

Increased opioid prescribing, particularly OA prescribing patterns associated with increased risk of overdose (i.e., high-dose, high-volume, and long-duration prescriptions),³ has been proposed as an important driver of non-medical OA use and OA-involved overdose deaths, with over 153 million OA prescriptions written nationally in 2019.⁴ Increased use of OAs to treat non-cancer pain occurred disproportionately in primary care compared with other specialties, such as oncology or pain management.⁵ As such, public health responses have targeted primary care in an effort to reduce OA prescribing patterns that may increase patient risk of overdose.⁶

Prescription drug monitoring programs (PDMPs), databases that track controlled substance prescribing and dispensing across patients and prescribers, are a primary policy strategy to reduce OA supply and prescribing that may increase overdose risk. While some PDMPs have existed for decades,⁷ the 2002 Harold Rogers Prescription Drug Monitoring Grant Program expanded federal support for PDMPs.⁸ As of 2020, all US states, Washington, DC, Puerto Rico, and Guam have active PDMPs.⁹

Operational details of PDMPs vary across jurisdictions, with voluntary or mandatory participation by prescribers and pharmacists depending on the state. As of July 2020, 46 states mandated prescriber PDMP use.¹⁰ PDMP mandates are associated with several positive outcomes, including reduced OA prescribing patterns associated with patient overdose risk,¹¹ opioid-related hospitalizations,¹² OA-associated overdose deaths,¹³ and injection drug use initiation.¹⁴

In New York State (NYS), the site of this study, the 2013 Internet System for Tracking Over-Prescribing (I-STOP) Act mandated prescriber PDMP consultation prior to prescribing schedule II, III, or IV medications.¹⁵ Research evaluating I-STOP identified consistency with mandatory PDMP use laws in other jurisdictions; I-STOP is associated with increases in prescriber PDMP usage,¹⁶ as well as decreases in aggregate OA prescribing as well as prescriptions issued by particular specialties.¹⁷

The impact of PDMP mandates on patient termination from chronic opioid therapy (COT) remains unclear. COT termination is an important secondary consideration given the risks of overdose and transition to heroin associated with lack of COT continuity.^18,19 Research on the risks of repeated COT discontinuation is necessary to identify patients who experience multiple terminations and concomitantly multiple periods of post-termination overdose risk. Several reasons for provider-initiated COT discontinuation are possible, including increased appropriate COT termination due to more frequent PDMP review or lack of clarity among providers about state or institutional PDMP use guidelines, as well as the potential for patient-initiated COT discontinuation. Emergent qualitative work has identified that some prescribers may use PDMP mandates to terminate patients from COT, contrary to federal and state PDMP guidelines.^20–22 Quantitative examination of these findings is crucial when considering possible reasons for COT discontinuation and developing public health education for providers about PDMP use.

This retrospective cohort study uses patient-level OA dispensation data to estimate the risk of “cycling” through COT episodes following the NYS PDMP mandate. We hypothesize that patients engaged in COT may experience increased cycling after the mandate, as patients with opioid dependence may seek COT elsewhere after termination. Given the risks of abrupt COT discontinuation, patients who experience multiple terminations may comprise a sub-population at elevated risk of overdose or transition to non-prescription opioids.

METHODS

Data Sources

Data came from IQVIA Longitudinal Prescription Data (LRx), a nationwide database of all-payer, patient-level OA prescriptions dispensed from outpatient retail pharmacies within the US.²³ Prescriptions obtained by mail or dispensed in long-term care facilities, as well as cough/cold formulations and methadone, were not included. From 2011 to 2015, the years of this study, IQVIA LRx national coverage rates ranged from 74 to 86%. Use of de-identified IQVIA LRx data is considered exempt by the New York University Grossman School of Medicine Institutional Review Board.

We defined OAs as schedule II and III formulations in the IQVIA LRx database excluding methadone. Buprenorphine formulations for pain were included; buprenorphine formulations for opioid use disorder (OUD) were excluded. Prescription dose was assessed using daily morphine milligram equivalents (MMEs), calculated for each prescription by dividing the overall MMEs by the days’ supply of the prescription. Prescriptions were considered high dose if daily MME exceeded 90, consistent with Centers for Disease Control and Prevention (CDC) opioid prescribing guidelines.⁶

Study Setting and Period

Data captured OAs dispensed between August 1, 2011, and July 31, 2015, in the five counties comprising NYC (Bronx, Kings, New York, Queens, and Richmond; referred to as “NYC”) and the six New Jersey counties directly bordering NYC (Bergen, Essex, Hudson, Middlesex, Monmouth, and Union; referred to as “NJ”). The unit of time for each prescription was month. Person-time was calculated using the number of months individuals engaged in COT during the study.

We restricted the intervention period to 2 years after the mandate enactment to capture short-term changes in OA prescribing associated with the mandate. Evidence indicates that OA prescriptions decline most markedly in the first year after PDMP implementation, with decreases sustained into the second and third years.¹⁰ As such, we are interested in changes in prescriptions issued to patients engaged in COT during a short-term transition period after mandate enactment.

We restricted the study setting to NYC, rather than across the whole of New York State (NYS), for two reasons. First, although the NYS Department of Health issued advance notification of the I-STOP enactment on its website and through press release, we are uncertain whether providers statewide were aware of the PDMP mandate in the ensuing 2 years. For providers in NYC, however, the NYC Department of Health and Mental Hygiene mailed a Dear Colleague letter to all licensed prescribers in NYC announcing the PDMP mandate shortly after I-STOP enactment.²⁴ In addition, the enactment of I-STOP received substantial local media coverage in NYC.²⁵ As such, we assume that awareness of I-STOP among providers in NYC is likely to have been more comprehensive than statewide. Second, given the difference-in-differences (DID) design, the restriction of the treatment jurisdiction to NYC and the control jurisdiction to the NJ counties allowed us to assume parallel OA prescribing trends prior to the mandate (Supplemental Figures 1, 2).

Study Measures

We defined COT as at least one OA prescription per month for three or more consecutive months, consistent with the CDC definition.⁶ One-month gaps in prescriptions were permitted, provided that patients at some time during their treatment tenure filled three consecutive months of prescriptions.

Eligibility was determined along the following criteria. First, all OA prescriptions dispensed in NYC and NJ from July 1, 2011, through August 31, 2015, were obtained from IQVIA LRx. Prescriptions were excluded if patients (1) were under age 18 at the start of the study; (2) were dispensed OA prescriptions in both NYC and NJ; and (3) engaged in only one episode of COT that terminated prior to the study endpoint. Age, sex, and prescription payment type were available for all patients.

We assessed one exposure: the August 2013 enactment of I-STOP, the NYS PDMP mandate. NYC was exposed; NJ was unexposed. We classified the 2-year period prior to August 2013 as unexposed time, and the 2-year period after August 2013 as exposed time. To our knowledge, no other policies differed between the two regions during the study period, including laws regulating the direct dispensing of controlled substances, laws regulating the operations of pain management clinics, requirements of commercial insurance and Medicaid coverage of medications to treat OUD, and guidelines for opioid prescribing for acute and emergency settings.²⁶

Our primary outcome was COT episodes, defined as a period of COT recurrence after COT cessation. All patients had at least one episode. We defined recurrence as a subsequent episode of COT after a first episode and cessation. We defined cessation as at least 2 months without an OA prescription after a period of COT. There was no upper limit to the number of COT recurrences. Patients were censored if their first COT episode continued unbroken to the study endpoint or if the study ended while they were engaged in COT.

Statistical Analysis

Analysis had three stages: descriptive statistics, hazard modeling, and sensitivity analysis. First, available baseline characteristics—including age, sex, and payment type—were tabulated for patients in the cohort. These were assessed for the full cohort and across episode strata.

Second, we estimated COT cycling risk after I-STOP enactment using Prentice, Williams, and Peterson gap-time (PWP-GT) models, a variant of Cox proportional hazard models that allow for the modeling of recurrent events.²⁷ The PWP-GT model includes individuals who experience k events into the risk set for the k + 1 event.²⁸ In our case, patients with one COT episode were included in the risk set for the second episode. Those with two COT episodes were included in the risk set for the third episode, representing a smaller selection of the cohort, and so on, until all episodes were exhausted. Failure time was calculated as the duration in months between COT episodes, reset to zero after each episode.

We applied PWP-GT models to a life-table DID design to estimate the average treatment effect on the treated, here the risk of patient COT cycling associated with enactment of I-STOP. Recent research has extrapolated DID designs for the valid estimation of causal effects for two-group, two-period cohorts for which the risk of an event (here, cycles of COT) varies with duration (here, time since the last cycle of COT).²⁹ This adaptation of DID to hazard processes allows for formal policy analysis with data generated by a continuous time process.³⁰ Assuming that the timing of a policy change (here, the enactment of I-STOP) is exogenous, then that exogeneity applies equally to a hazard difference-in-difference design. As such, the estimation of a two-group, two-period hazard DID model differs from a standard two-group, two-period linear probability DID model only in that the outcome is a hazard process.

We adjusted models for patient age group, a dummy variable indicating sex, prescription payment method (private insurance, cash, Medicaid, or Medicare), and a dummy variable indicating high-dose prescriptions. To account for variation in rurality between the counties included in this study, which prior research has shown to be associated with both clinician prescribing and patient OA use,^31,32 we adjusted models for county population density. Population density for all counties were obtained from the US Census/American Community Survey 2014–2018 estimates.³³

We visually tested for parallel trends in OA prescriptions in NYC and NJ consistent with the modeling assumptions necessary for DID; this visual inspection indicated that prescribing rates were parallel during the study period (Supplemental Figures 1, 2). As such, we cautiously interpret the DID estimator here as the causal hazard ratio of cycling due to implementation of I-STOP among patients engaged in COT in NYC.

To assess the potential differential cycling risk among patients receiving high-dose COT, we conducted a planned subgroup analysis among patients receiving high-dose prescriptions. The model for this analysis took the same form as above, and was adjusted for age group, sex, payment type, and county population density.

Finally, we conducted a sensitivity analysis using only patients’ first episodes of cycling, ignoring subsequent cycles. This was to test the robustness of the PWP-GT DID model to higher volumes of cycling. This model took the form of a Cox proportional hazard model fit within the DID framework described above.²⁵ For this model, failure time remained the duration between episodes, as the first episode marked a given patient’s entrance into the study. We modeled the overall and high-dose subgroup-specific hazards as part of this sensitivity analysis. All analyses were conducted using Stata version 15.1 (StataCorp LLC, College Station, TX).

RESULTS

Sample Characteristics

The cohort comprised 7604 patients, and dispensed a total of 12,695 OA prescriptions. Of prescriptions dispensed in NJ (n=6447), 42.0% were dispensed to patients aged 35 to 54, 53.3% were dispensed to female patients, 63.5% were paid using private insurance, 25.5% were high-dose, and 96.6% were dispensed during patients’ first, second, or third COT episodes. The mean change in daily MME between episodes was an increase of 5.52 MME per day. The mean population density across the NJ counties was 1914.4 persons per square kilometer. Prescriptions censored at the study endpoint represented 2.0% of the NJ total (Table 1).

Table 1.

Characteristics of Prescriptions Dispensed to Patients in New York City and New Jersey, August 1, 2011–July 31, 2015

	New Jersey	New York City	Overall	P-value
	N (%)	N (%)	N (%)
	6447 (100)	6178 (100)	12,625 (100)
Age group
18–34	792 (12.3)	603 (9.8)	1395 (11.1)	< 0.001
35–54	2708 (42.0)	2389 (38.7)	5097 (40.4)
55–64	1523 (23.6)	1681 (27.2)	3204 (25.4)
65+	1424 (22.1)	1505 (24.4)	2929 (23.2)
Gender
Female	3435 (53.3)	3397 (55.0)	6832 (54.1)	0.055
Male	3012 (46.7)	2781 (45.0)	5793 (45.9)
Payment type
Private insurance	4096 (63.5)	3839 (62.1)	7935 (62.9)	< 0.001
Cash	603 (9.4)	343 (5.6)	946 (7.5)
Medicaid	86 (1.3)	137 (2.2)	223 (1.8)
Medicare	1662 (25.8)	1859 (30.1)	3521 (27.9)
Daily MME change
Mean (SD)	5.5 (0.7)	4.8 (0.7)	5.2 (0.5)	< 0.001
High-dose
High-dose prescriptions	1642 (25.5)	1546 (25.0)	3188 (25.3)	0.565
Population density
Mean (SD)	1914.4 (1488.0)	14,943.7 (8,354.0)	8290.2 (78.5)	< 0.001
COT cycles
1	3921 (60.8)	3680 (59.6)	7601 (60.2)	0.404
2	1704 (26.4)	1680 (27.2)	3384 (26.8)
3	606 (9.4)	612 (9.9)	1218 (9.7)
4	163 (2.5)	170 (2.8)	333 (2.6)
5	43 (0.7)	30 (0.5)	73 (0.6)
6	9 (0.2)	6 (0.1)	15 (0.1)
7	1 (0.02)	0 (0)	1 (0.01)
Censored
Censored prescriptions	130 (2.0)	145 (2.4)	275 (2.2)	0.203

Notes: Percentages may not sum to 100 due to rounding; chi² comparing characteristics among treated (NYC) to control (NJ) prescriptions; daily MME change denotes the mean change in MME/day between episodes of COT; NJ captures prescriptions dispensed in Bergen, Essex, Hudson, Middlesex, Monmouth, and Union counties; NYC captures prescriptions dispensed in Bronx, Kings, New York, Queens, and Richmond counties; population density reflects persons per square kilometer

Source: IQVIA LRx

Of prescriptions dispensed to NYC patients (n=6178), 38.7% were dispensed to patients aged 35 to 54, 55.0% were dispensed to female patients, 62.1% were paid using private insurance, 25.0% were high-dose, and 96.7% were dispensed during patients’ first, second, or third COT episodes. The mean change in daily MME between episodes was an increase of 4.83 MME per day. The mean population density across the NYC counties was 14,943.7 persons per square kilometer. Prescriptions censored at the study endpoint represented 2.4% of the NYC total (Table 1).

The number of COT episodes ranged from 1 to 7 (median=1; IQR=1–2). One patient experienced the maximum number of episodes. Most prescriptions across episodes were dispensed to patients aged 35 to 54 or 55 to 64. Female patients were dispensed a higher proportion of prescriptions across episodes than male patients, and most prescriptions across episodes were paid using private insurance. The proportion of high-dose prescriptions increased with the number of episodes. The mean change in daily MME between episodes was highest during the transition from third to fourth episode (+ 8.1 MME per day) and lowest during the transition from fifth to sixth episode (–12.9 MME per day). Mean population density remained relatively stable across episodes. The proportion of censored prescriptions increased with the number of episodes (Table 2).

Table 2.

Characteristics of Prescriptions Dispensed by Episode of Chronic Opioid Therapy, August 1, 2011–July 31, 2015

	Chronic opioid therapy episode
	1	2	3	4	5	6	7
	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)
	7601 (100)	3384 (100)	1218 (100)	333 (100)	73 (100)	15 (100)	1 (100)
Age group
18–34	937 (12.3)	335 (9.9)	92 (7.6)	28 (8.4)	2 (2.7)	1 (6.7)	0 (0)
35–54	3185 (41.9)	1329 (39.3)	445 (36.5)	113 (33.9)	19 (26.0)	5 (33.3)	1 (100)
55–64	1869 (24.6)	884 (26.1)	337 (27.7)	88 (26.4)	22 (30.1)	4 (26.7)	0 (0)
65+	1610 (21.2)	836 (24.7)	344 (28.2)	104 (31.2)	30 (41.1)	5 (33.3)	0 (0)
Gender
Female	4081 (53.7)	1858 (54.9)	664 (54.5)	183 (55.0)	37 (50.7)	9 (60)	0 (0)
Male	3520 (46.3)	1526 (45.1)	554 (45.5)	150 (45.1)	36 (49.3)	6 (40)	1 (100)
Payment type
Private insurance	4820 (63.4)	2103 (62.2)	740 (60.8)	217 (65.2)	45 (61.6)	9 (60)	1 (100)
Cash	659 (8.7)	213 (6.3)	65 (5.3)	7 (2.1)	3 (4.1)	0 (0)	0 (0)
Medicaid	151 (2.0)	59 (1.7)	9 (0.7)	3 (0.9)	1 (1.4)	0 (0)	0 (0)
Medicare	1972 (25.9)	1009 (29.8)	404 (33.2)	106 (31.8)	24 (32.9)	6 (40)	0 (0)
Daily MME change
Mean (SD)	5.9 (0.7)	4.2 (0.9)	3.1 (1.6)	8.1 (3.3)	2.3 (6.9)	-12.9 (13.1)	-7.3 (.)
High-dose
High-dose prescriptions	1,919 (25.3)	869 (25.7)	290 (23.8)	82 (24.6)	22 (30.1)	6 (40)	0 (0)
Population density
Mean (SD)	8,246.5 (8,792.4)	8,407.0 (8,883.4)	8,319.3 (8,765.9)	8,171.2 (8,886.0)	7,523.9 (8,533.0)	8,524.4 (9,670.1)	2,430.1 (0)
Censored
Censored prescriptions	5 (0.1)	120 (3.6)	96 (7.9)	37 (11.1)	15 (20.6)	2 (13.3)	0 (0)

Notes: Percentages may not sum to 100 due to rounding; daily MME change denotes the mean change in MME/day between episodes of COT; NJ captures prescriptions dispensed in Bergen, Essex, Hudson, Middlesex, Monmouth, and Union counties; NYC captures prescriptions dispensed in Bronx, Kings, New York, Queens, and Richmond Counties; population density reflects persons per square kilometer across NYC and NJ

Source: IQVIA LRx

Risk of Chronic Opioid Therapy Cycling

Results from the PWP-GT models indicated that, for the full cohort, the adjusted hazard ratio for COT cycling for patients in NYC relative to patients in NJ after the implementation of I-STOP was 1.01 (95% CI, 0.94–1.08). Among the subgroup of patients receiving high-dose prescriptions in NYC relative to high-dose patients in NJ, the estimated risk of cycling was 1.16 (95% CI, 1.01–1.34) (Table 3). Sensitivity analysis conducted using a Cox proportional hazards model indicated that the risk of having a single recurrence of COT for patients in NYC relative to patients in NJ after the implementation of I-STOP was 1.01 (95% CI, 0.94–1.11). Among the subgroup of patients receiving a high-dose prescription in NYC relative to high-dose patients in NJ, the estimated risk of cycling toward a single recurrence was 1.09 (95% CI, 0.91–1.31) (Table 4).

Table 3.

Difference-in-Differences Hazard Estimates for Patient Recurrence of Chronic Opioid Therapy After the Enactment of I-STOP, August 1, 2011–July 31, 2015: Prentice, Williams, and Peterson Gap-Time Model

	All prescriptions (n = 12,625)
	HR (95% CI)	Person-months at risk
Unadjusted	1.01 (0.94, 1.08)	88,785
Adjusted	1.01 (0.94, 1.08)	88,785
	High-dose prescriptions (n = 3188)
	HR (95% CI)	Person-months at risk
Unadjusted	1.17 (1.02, 1.34)	21,930
Adjusted	1.16 (1.01, 1.34)	21,930

Notes: All prescription models adjusted for age group, sex, payment method, high-dose prescriptions, and county population density; high-dose prescription model adjusted for age group, sex, payment method, and county population density

Source: IQVIA LRx

Table 4.

Difference-in-Differences Hazard Estimates for Single Recurrence of Chronic Opioid Therapy After the Enactment of I-STOP, August 1, 2011–July 31, 2015: Cox Proportional Hazards Model

	All prescriptions (n = 7601)
	HR (95% CI)	Person-months at risk
Unadjusted	1.01 (0.92, 1.11)	49,706
Adjusted	1.01 (0.92, 1.11)	49,706
	High-dose prescriptions (n = 1919)
	HR (95% CI)	Person-months at risk
Unadjusted	1.09 (0.91, 1.31)	12,211
Adjusted	1.09 (0.90, 1.31)	12,211

Notes: All prescription models adjusted for age group, sex, payment method, high-dose prescriptions, and county population density; high-dose prescription model adjusted for age group, sex, payment method, and county population density

Source: IQVIA LRx

DISCUSSION

This study estimated the impact of the NYS PDMP mandate on patient risk of COT cycling and identified that the mandate enactment had no overall effect. However, the mandate increased cycling risk for patients receiving high-dose COT by 16%. This increased risk attenuated in a sensitivity analysis restricting estimates to the transition from the first to second COT episodes. Overall, findings suggest that patients receiving high-dose OAs with multiple COT episodes were at increased risk of cycling following the mandate enactment.

As states expand and renew PDMP mandates, public health agencies should bolster provider education on COT and OUD. Prior research has shown that PDMP mandates are associated with increased provider use of PDMP systems.³⁴ It is possible that more frequent PDMP use may lead to increased COT discontinuation, as providers become increasingly aware of patients with potential OUD. Future research should consider the frequency of provider PDMP use as a factor in patient COT discontinuation. Furthermore, providers may experience a lack of clarity around PDMP use guidelines, particularly for providers receiving mixed guidance from public health authorities and institutional oversight figures.³⁵ Provider public health education could clarify discrepancies and offer appropriate strategies for COT termination when appropriate.

Additionally, research indicates that patients are at increased risk of fatal overdose following COT discontinuation¹⁹; as such, patients who cycle through COT may constitute a high-risk group. Additional work has identified that the patients at greatest risk of discontinuation are those receiving high-dose OAs and co-prescriptions with benzodiazepines.³⁶ Our findings that patients receiving high-dose prescriptions were most likely to experience multiple COT cycles align with prior work.

Although COT discontinuation and recurrence remain understudied, one study of Oregon Medicaid beneficiaries identified that over 37% of COT patients experienced an abrupt discontinuation.³⁷ Contextualized with evidence that mortality risk is highest immediately after discontinuation,³⁸ our findings may suggest that patients who experience increased cycling—and discontinuation—may be prone to adverse outcomes and may warrant targeted health interventions.

Notably, few patients had high-frequency cycling; most patients had one episode and nearly all patients had three or fewer episodes. This suggests that most patients, upon COT discontinuation, did not engage in future COT. Our data are unable to indicate what might happen to those patients. While speculative, it is possible that patients may have chosen to end COT, died from overdose or another cause, or transitioned to non-prescription opioids. For the small group of patients who repeatedly enter and discontinue high-dose COT, tailored interventions, such as naloxone co-prescribing³⁹ or increased provision of buprenorphine in primary care,⁴⁰ may mitigate possible adverse outcomes associated with COT discontinuation.

Limitations

This study has several limitations. First, IQVIA LRx contains data on dispensed prescriptions, but lacks prescriber details and information on how or by whom medications were used. Our study assumed that patients were chronically taking opioids for pain. However, it is possible that included prescriptions reflected use for other indications such as cancer or dental treatments, both of which are separate from COT for chronic pain and may plausibly indicate cycling. Second, recurrent event modeling ignored patients who discontinued COT without recurrence. It is possible that these patients represent a separate risk group whose termination might influence that of other patients. That is, prescribers may identify COT patients with OA histories characteristic of increased overdose risk through mandated PDMP use and terminate such patients. Should those terminated patients not seek future COT, then the patients in our cohort may be at reduced likelihood of COT engagement or increased likelihood of COT termination from similar prescribers. Third, all patients initiated COT during the observation period, but we are uncertain whether these represent patients’ first COT episodes. It is possible that some patients cycled prior to the study. However, as nearly all patients in our sample experienced three or fewer cycles during a 4-year period, we assume that left censoring is not a major concern.

Fourth, prescriptions are linked by patients, not prescribers. Thus, we are unable to assess whether patients engaged in subsequent episodes of COT with the same or different prescribers. Fifth, other unmeasured events such as provider education campaigns or law enforcement activities, may have contributed to increased cycling. To our knowledge, no jurisdiction-wide prevention or enforcement activities occurred in NYC near the August 2013 PDMP mandate. Sixth, differences in opioid policies between NYC and NJ may have contributed to cycling. However, no differences during the study period were identified for the following laws and policies: laws regulating the direct dispensing of controlled substances; laws regulating the operations of pain management clinics; coverage requirements of commercial insurance and Medicaid for medications to treat OUD; and guidelines for opioid prescribing in acute and emergency settings.²⁶

Seventh, our data source did not capture several potential patient-level confounders. We only were able to adjust models for patient age group, sex, and prescription dose. Other covariates, such as patient education, medical comorbidities, and prescriber specialty, were not available, which may introduce bias into this study. Eighth, our data source did not capture patient death. Thus, we are unable to identify whether patients discontinued COT episodes due to death, a limitation that may underestimate the proportion of censored observations.

CONCLUSIONS

This study examined the risk of cycling through COT associated with the enactment of I-STOP, the NYS law mandating prescriber PDMP use. We found no overall change in COT cycling after the PDMP mandate enactment in NYS. However, our findings show that enactment of the NYS PDMP mandates increased the risk of COT cycling among patients with high-dose OA prescriptions. Future research is needed to assess associations between patient COT cycling and overdose risk, secondary outcomes of patient COT discontinuation after PDMP mandates (e.g., OUD treatment initiation or transition to heroin or other illicit opioids), and provider decision making processes in response to information gleaned through PDMP use.

Funding

This study was supported by the Center for Opioid Epidemiology and Policy in the Department of Population Health, New York University Grossman School of Medicine. BA received support from the National Institute on Drug Abuse (T32 DA007233-37).

Declarations

Conflict of Interest

The authors declare that they do not have a conflict of interest.