Trajectories of opioid prescribing by general dentists, specialists, and oral and maxillofacial surgeons in the United States, 2015–2019

Abstract

Background.

Despite decreases in opioid prescribing from 2016 through 2019, some dentists (general, specialists, oral and maxillofacial surgeons) in the United States continue to prescribe opioids at high rates. The authors’ objective was to define dentists’ trajectories of opioid prescribing.

Methods.

The authors identified actively prescribing dentists from the IQVIA Longitudinal Prescription data set, from 2015 through 2019. Group-based trajectory modeling identified opioid prescribing trajectories on the basis of dentists’ annual prescribing rates for the overall sample (model 1) and for high prescribers (model 2). The authors used χ² or Mann-Whitney U tests to characterize the model 2 trajectory groups.

Results.

In model 1 (n = 199,145 prescribers), group-based trajectory modeling identified 8 trajectories that were grouped into 5 categories. A total of 14.8% were nonprescribers who composed less than 1% of all prescriptions, low prescribers (3 groups; 46.0%) prescribed at low rates (2015: 5.5%–16.9%; 2019: 1.5%–11.9%), decliners (7.3%) decreased prescribing rapidly (2015: 29.4%; 2019: 5.1%), moderately high prescribers (2 groups; 28.5%) prescribed moderately (2015: 28.7% and 39.2%; 2019: 18.1% and 28.8%), and consistently high prescribers (3.4%) prescribed at high rates (2015: 54.6%; 2019: 44.7%). In model 2, from consistently high prescribers (n = 6,845), 4 trajectories were identified. Of these 4 groups, 1 group (7.5%) declined prescribing rapidly. The groups did not differ meaningfully; however, the rapid decliners included fewer oral and maxillofacial surgeons (13.0% vs 18.4%), saw more Medicaid patients (2.5% vs 1.0%), and had higher opioid prescribing rates in 2015 (95.5% vs 91.6%) (P < .001 for all).

Conclusions.

The authors identified variations in dentists’ opioid prescribing rates. Although 60% of dentists decreased prescribing rates by 30% through 83%, 3.4% of dentists consistently prescribed at high rates.

Practical Implications.

Some dentists continue to prescribe opioids at high levels, indicating that additional information is needed to better inform policy and clinical decision making.

The United States is dealing with a deadly opioid epidemic with huge economic and societal impact.^1–3 In response, the Centers for Disease Control and Prevention (CDC), the American Dental Association, and several state dental boards recommend nonsteroidal anti-inflammatory drugs (NSAIDs) as the first-line therapy to manage acute dental pain.^4–6 However, compared with prescribing in other countries, opioid prescribing by US dentists is common and excessive.^7,8 Furthermore, opioid prescriptions by dentists have been associated with serious adverse outcomes including abuse, persistent opioid use, and diversion.^8–12 Nationally and at the states level, efforts have been made to decrease opioid prescribing, such as instituting prescriptions drug monitoring programs (PDMPs) and requiring continuing education (CE) courses on dental pain management for licensure and relicensure. The most recent national data available show that overall opioid prescribing by dentists and oral and maxillofacial surgeons (OMFSs) has decreased by 41% and 20%, respectively, from 2016 through 2019.^13,14 Despite the overall decreasing trend, variation exists in prescribing levels by dentists, with some geographic areas, for example, experiencing an increase in prescribing.^13–15 This variation implies that some dentists continue to prescribe opioids for dental pain management and are not following evidence-based recommendations.¹⁶ However, who those dentists are and how they differ from those who have decreased their prescribing is not well understood. Without this information, interventions cannot be tailored to address the barriers to decreasing rates faced by dentists who continue to prescribe opioids at high rates.

The objective of this study was to identify trajectories of US dentists from a longitudinal prescription data set from 2015 through 2019 and to characterize high prescribers using group-based trajectory modeling (GBTM). GBTM is a method that identifies subgroups that have followed a similar trajectory on the basis of an outcome progression over time.¹⁷ GBTM provides a graphical representation of the subgroup trajectories that can be visually and statistically compared.¹⁷ In the context of opioid prescribing, GBTMs have been used mainly to describe patient-related outcomes such as overdose risk and persistent opioid use and have not been used widely to identify trajectories of opioid prescribing by providers.^18–20

We expected to identify at least 3 groups in our analysis: (1) a group of dentists who consistently prescribed opioids at low rates over the course of the study, (2) a group of dentists who were high opioid prescribers at baseline (ie, 2015) and who continued to prescribe opioids at high rates over the course of the study, and (3) a group of dentists who were high prescribers at baseline and who decreased their prescribing over the course of the study. To our knowledge, this is the first study that identifies opioid prescribing trajectories in dentistry using a longitudinal national prescribing data set. Comparing the characteristics of consistently high opioid prescribers with those who decreased prescribing will enable the field to develop targeted interventions to reduce opioid prescribing among US dentists who have not responded already to previous interventions.

METHODS

We conducted a cohort study using IQVIA Longitudinal Prescription Data (LRx)²¹ from 2015 through 2019. We refer to 2015 as the baseline and 2019 as the end of the study period. LRx encompasses approximately 92% of all dispensed outpatient prescriptions in the United States, including cash payments. Prescriber data included dentist provider types grouped into general dentist, specialist (pediatric, endodontist, orthodontist, periodontist, prosthodontist), and OMFS (dental anesthesiologist, OMFS). We refer to these groups of dental providers collectively as dentists.

As done previously,²² we included actively prescribing dentists defined as those with 20 or more prescriptions of any medication in a calendar year within LRx. Opioid prescriptions were identified using National Drug Codes. As done previously,¹³ we excluded topical opioids, opioid formulations used exclusively as a cough suppressant (eg, promethazine and codeine), or buprenorphine-containing products used for opioid use disorder treatment and not indicated for pain (eg, buprenorphine and naloxone). In addition, we excluded prescriptions with missing dentist, patient, or prescription information and prescriptions with days’ supply or quantity dispensed that were 0 or less or above the 99th percentile (numbers < 1 are usually entry errors and were small in number). We excluded dentists who were not within the 50 states and District of Columbia and prescriptions by dental hygienists or technicians (eg, fluoride, chlorohexidine, and other topical medications) (eFigure 1, available online at the end of the article).

We grouped geographic location into US Census Bureau regions (Northeast, Midwest, South, West) based on dentist location. For each dentist, we calculated a patient-based prescribing rate defined as the total number of prescriptions of any medication per 100 unique patients per dentist per year. Similarly, we calculated a patient-based opioid prescribing rate as the total number of opioid prescriptions per 100 patients per dentist per year. We summarized the patient-based prescribing measures as the median and 25th percentile through 75th percentile (Q1-Q3) number of prescriptions each prescriber wrote per patient annually.

Patient data included age groups (children, 0-< 12 years; adolescents, 12-< 18 years; adults, 18-< 65 years; older adults, 65 years); sex (female, male); and payer (cash, third party, Medicaid, Medicare). We determined each provider’s patient mix by calculating the proportion of each categorical variable per dentist per year.

Opioid prescription information included strength, dosage form, quantity dispensed, days’ supply, and date dispensed. We calculated opioid dosages in morphine milligram equivalents (MMEs) (total daily dose × opioid strength × MME conversion factor). We identified proportion of prescriptions greater than 3 days’ or 7 days’ supply and opioid dosages that are defined as high risk per CDC guidelines (≥50 or ≥90 MME per day).²³ Annually, for each provider, we calculated the proportion of prescriptions that were defined as high risk.

Our outcome measure was the opioid prescribing rate for each dentist. Using provider-level data on all dental prescriptions in LRx, we calculated the opioid prescribing rate as the number of opioid prescriptions divided by the total number of prescriptions (of any medication) for a specific dentist in a year and converted to rate by multiplying by 100. This measure enabled us to determine changes in opioid prescribing behavior by comparing opioid prescriptions by a dentist to his or her overall prescribing of any medication. This outcome of annual opioid prescribing rate per dentist was used to define trajectories in our GBTM.^24,25

We conducted 2 separate GBTM analyses. Model 1 included all dentists in our cohort to describe trajectories of opioid prescribing rates. Model 2 included only dentists in the highest group of opioid prescribing identified in model 1, with the objective of characterizing factors associated with variation in opioid prescribing within this high group. Details on model selection and fitting can be found in the Appendix, available online at the end of this article.

For each of the groups identified by the GBTM of model 1 and 2, we calculated frequencies and percentages of dentists’ characteristics (ie, dentist type and geographic region). Moreover, for each group trajectory identified, we describe patient mix and opioid prescription aggregated at the group level and presented as medians (Q1-Q3). For groups identified in model 2 (ie, trajectories within high prescribers only), we used Mann-Whitney U tests to compare medians for continuous variables and χ² test to compare frequencies for categorical variables.

We performed data analyses using SAS Version 9.4 (SAS Institute) with the PROC TRAJ plugin for SAS to create the trajectories.^26,27

RESULTS

From 2015 through 2019, there were 319,384,935 prescriptions by 199,145 unique dentists, of which 70,978,575 (22.2%) were for opioids. Over the 5 years examined in our study, the number of opioids prescribed by dentists decreased by 34.4% (2015, 16.9 million; 2019, 10.9 million; Figure 1). At baseline (2015), 95% of prescribers of opioids were general dentists, 1.6% were specialists, and 3.4% were OMFSs (Table 1). The median (Q1-Q3) patient-based prescribing rate for any medication was 160.7 (141.1–186.1) per 100 unique patients per dentist per year, and the median (Q1-Q3) patient-based opioid prescribing rate was 31.8 (13.3–58.1) per 100 unique patients per dentist per year. Overall, 4.2% of opioid prescriptions per provider were 50 or more MME per day and 28.6% were over 3 days’ supply.

Figure 1.

All prescriptions and opioid prescriptions by actively prescribing general dentists, specialists, and oral and maxillofacial surgeons, from 2015 through 2019.

Table1.

Characteristics of the cohort in 2015 based on model 1 group membership.^*

GROUPS	ALL GROUPS (N = 199,145)	NONPRESCRIBERS (n = 29,530, 14.8%)	LOW PRESCRIBERS (n = 91,511, 46.0%)	DECLINERS (n = 14,491, 7.3%	MODERATELY HIGH PRESCRIBERS (n = 56,768, 28.5%)	CONSISTENTLY HIGH PRESCRIBERS (n = 6,845, 3.4%)
Provider Type, No. (%)
General dentist	189,849 (95.3)	28,436 (96.3)	89,942 (98.3)	14,106 (97.3)	51,813 (91.3)	5,552 (81.1)
Oral and maxillofacial surgeon	6,734 (3.4)	147 (0.5)	596 (0.7)	252 (1.7)	4,505 (7.9)	1,234 (18.0)
Specialist	2,562 (1.3)	947 (3.2)	973 (1.1)	133 (0.9)	450 (0.8)	59 (0.9)
Provider Region, No. (%)
Northeast	39,498 (19.8)	11,203 (37.9)	20,818 (22.8)	2,523 (17.4)	4,513 (8.0)	441 (6.4)
Midwest	40,012 (20.1)	4,961 (16.8)	19,918 (21.8)	3,145 (21.7)	10,610 (18.7)	1,378 (20.1)
South	66,505 (33.4)	6,995 (23.7)	26,211 (28.6)	5,039 (34.8)	24,990 (44.0)	3,270 (47.8)
West	53,130 (26.7)	6,371 (21.6)	24,564 (26.8)	3,784 (26.1)	16,655 (29.3)	1,756 (25.7)
Patient Mix,† Median (Q1-Q3‡)
Patient-based 11 prescribing rate§	50.7 (141.1–186.1)	151.7 (129.4–183.9)	158.3 (140.3–181.5)	158.7 (141.9–180.3)	169.7 (149.4–196.6)	156.0 (138.9–177.0)
Payer, %
Medicare	13.4 (7.6–21.2)	12.1 (1.6–25.8)	15.8 (9.0–24.1)	12.2 (7.5–18.0)	11.7 (7.5–17.1)	10.7 (7.0–15.0)
Cash	10.8 (6.4–16.9)	9.3 (4.6–16.9)	10.5 (6.3–16.5)	10.9 (6.7–16.7)	11.6 (7.4–17.3)	12.3 (7.5–18.6)
Third party	70.3 (60.4–78.5)	70.0 (56.5–81.4)	69.0 (59.8–77.1)	71.7 (62.1–79.2)	72.0 (63.2–79.1)	70.4 (58.9–78.8)
Medicaid	0.0 (0.0–2.2)	0.0 (0.0–2.1)	0.0 (0.0–1.6)	0.2 (0.0 −3.4)	0.3 (0.0–2.9)	1.1 (0.0 −7.4)
Age group, %
Children	0.7 (0.0–2.7)	1.2 (0.0 −16.3)	0.7 (0.0–2.7)	0.8 (0.0 −2.4)	0.7 (0.0 −1.9)	0.6 (0.0–1.6)
Adolescent	1.4 (0.0–3.7)	1.7 (0.0–7.4)	1.2 (0.0–3.0)	1.5 (0.3 −3.3)	1.7 (0.5–4.0)	2.2 (0.7–5.9)
Adult	75.0 (61.7–84.2)	57.1 (26.1–77.9)	71.3 (58.8–81.7)	80.0 (71.0–86.8)	80.0 (72.2–86.5)	83.0 (75.1–88.9)
Older adult	17.4 (8.6–29.8)	15.8 (2.3–38.0)	22.4 (11.5–35.0)	14.4 (7.7–23.3)	14.2 (8.3–22.0)	11.1 (6.4–16.7)
Opioid Prescribing,† Median (Q1-Q3‡)
Patient-based opioid prescribing rate	31.8 (13.3–58.1)	2.7 (1.5–4.3)	17.5 (9.6–27.8)	48.1 (38.3–60.4)	62.0 (47.7–77.9)	91.9 (81.5–101.6)
High-risk prescribing, %
Morphine milligram equivalent ≥ 50/d	4.2 (0.0–19.2)	0.0 (0.0–0.0)	1.9 (0.0–16.7)	5.0 (0.0–17.2)	7.5 (1.3–23.7)	10.0 (2.4–30.2)
Morphine milligram equivalent ≥ 90/d	0.0 (0.0–0.0)	0.0 (0.0–0.0)	0.0 (0.0–0.0)	0.0 (0.0–0.4)	0.0 (0.0–1.3)	0.3 (0.0–1.8)
> 3 d	28.6 (9.2–61.1)	33.3 (0–100)	28.6 (8.3–61.9)	27.4 (10.2–57.6)	28.7 (11.4–58.6)	26.3 (10.2–56.5)
> 7 d	0.0 (0.0–1.5)	0.0 (0.0–0.0)	0.0 (0.0–1.2)	0.0 (0.0–1.7)	0.3 (0.0–1.7)	0.5 (0.0–1.5)

^*Because medians and 25th and 75th percentiles are used, the totals may not add to 100%.

^†Patient mix and opioid prescribing measures were calculated at the unique dentist level and then aggregated to the group level.

^‡Q1-Q3: 25th percentile-75th percentile.

^§Includes all prescriptions (opioids and nonopioids).

Trajectories and declines in prescribing among all dentists (model 1)

From our GBTM analysis including all 199,145 dentists, we selected the 8-group linear model on the basis of our model selection criteria. Steps to select the model are described in eFigure 2 and eTables 1 and 2, available online at the end of this article.

We further categorized the 8 group trajectories into 5 categories on the basis of similar trends of opioid prescribing: nonprescribers, low prescribers, decliners, moderately high prescribers, and consistently high prescribers (Figure 2; Table 1; eTable 3, available online at the end of this article). Nonprescribers included 14.8% of the dentists, and this group’s opioid prescribing rates were consistently below 1% of total prescriptions. Low prescribers consisted of 3 groups (15.2%, 17.7%, and 13.1% of the dentists) who had similar trajectories and prescribed opioids at low rates throughout the study. The decliner group was composed of 7.3% of the dentists and included those who prescribed opioids at moderately high rates at baseline and who rapidly declined in prescribing over the study (82.7% decrease). The moderately high prescribers consisted of 2 groups (16.4% and 12.1% of dentists) with similar trajectories. Although this group showed a decline, they prescribed opioids at moderately high rates throughout the study. Finally, a small percentage of dentists (3.4%) displayed consistently high opioid prescribing rates with only a slight decrease in prescribing (18.0% decrease). Because the trajectories were defined on the basis of dentists’ opioid prescribing rates in 2015, we described the 5 groups’ baseline characteristics (Table 1).

Figure 2.

Model 1: Trajectories of opioid prescribing rates among dentists and percentage of prescribers in each trajectory, from 2015 through 2019 (n = 199,145).

Trajectories and characteristics of the consistently high prescribing group only (model 2)

For model 2, GBTM analysis of the consistently high prescribing group (n = 6,845), we selected a 4-group linear model (model selection described in eTables 4 and 5, available online at the end of this article). Figure 3 displays the trajectories identified in model 2. Among the 4 trajectories within the consistently high prescribing groups, 1 group (7.5% of prescribers) showed a rapid decline in opioid prescribing, starting with an opioid prescribing rate of 66.0% of total prescriptions in 2015 and declining to 26.5% of total prescriptions in 2019, a 59.9% decrease. Due to the rapid decrease in opioid prescribing, we refer to this group as rapid decliners. Although this group’s opioid prescribing rate was the second highest of the 4 at baseline, by 2019 it had the lowest prescribing rate within the consistently high prescribing group. The remaining 3 groups showed similar flat trends in opioid prescribing, decreasing by only 11.7%, 14.0%, and 8%. As these 3 groups had similar trends that contrasted with rapid decliners, we grouped them together and refer to them collectively as the combined other groups (eTable 6).

Figure 3.

Model 2: Trajectories of opioid prescribing rates among dentists within the consistently high opioid prescribing group identified in model 1 (n = 6,845).

In exploratory analyses, we compared the characteristics of the rapid decliners and the combined other groups:

• Provider type. The rapid decliners had greater proportions of both general dentists and specialists than the combined other groups, which had a higher proportion of OMFSs (P = .005). There were no differences by geographic region (Table 2).

• Patient mix. Table 3 describes the patient mix of decliners compared with the combined other groups at baseline and at the end of the study, because these characteristics changed over time. At baseline, rapid decliners received a greater proportion of reimbursements from Medicaid compared with the combined other groups (P < .001). By 2019, the rapid decliners and the combined other groups both decreased to similar lower levels of Medicaid reimbursements. At baseline, compared with the combined other groups, the rapid decliners wrote more total prescriptions to adults (P < .001) and fewer prescriptions to adolescents and older adult patients (P = .003 and P < .001, respectively). By 2019, the rapid decliners and the combined other groups both had shifted their prescribing such that adults made up a smaller percentage and older adults made up a larger percentage of their total prescribing, with rapid decliners writing fewer total prescriptions to adults than the combined other groups.

• Opioid prescribing. In 2015, the median patient-based opioid prescribing rate was higher among rapid decliners than the combined other groups (P < .001). By 2019, the rapid decliners had a significantly lower median patient-based opioid prescribing rate than the combined other groups. The rapid decliners had significantly lower rates of prescribing among 3 of the 4 high-risk prescribing indicators at baseline and all 4 by 2019.

Table 2.

Provider characteristics of model 2 groups (consistently high prescribers only).^*

GROUP	RAPID DECLINERS	COMBINED OTHER GROUPS	P VALUE†
Provider Type, %
General dentist	85.4	80.8	.005
Oral and maxillofacial surgeon	13.0	18.4
Specialist	1.6	0.8
Provider Region, % ‡
Northeast	6.5	6.4	.911
Midwest	21.3	20.1
South	46.3	47.9
West	25.9	25.9

^*Total N = 6,845; decliners, n = 513; all other groups, n = 6,332.

^†P values derived from χ² test.

^‡Total may not add up to 100% due to rounding.

Table 3.

Characteristics of model 2 groups (consistently high prescribers only) at baseline and at the end of the study.^*,†

GROUPS	BASELINE (2015)			END OF STUDY (2019)
	Rapid Decliners	Combined Other Groups	P Value‡	Rapid Decliners	Combined Other Groups	P Value‡
Patient Mix,§ Median (Q1-Q3¶)
Patient-based prescribing rate#	138.0 (125.4–151.6)	157.5 (140.3–178.2)	< .001	142.6 (127.5–163.0)	157.6 (141.8–176.6)	< .001
Payer, %
Medicare	9.9 (6.2–14.4)	10.7 (7.1–15.1)	.026	14.9 (9.4–21.7)	14.1(9.8–19.4)	.240
Cash	12.9 (7.4–20.0)	12.2 (7.5–18.5)	.302	9.7 (5.5–15.8)	8.7 (5.4–13.5)	.037
Third party	68.0 (50.9–78.3)	70.6 (59.3–78.8)	.001	68.7 (54.6–78.6)	72.7 (62.4–79.9)	< .001
Medicaid	2.5 (0.0–13.2)	1.0 (0.0–7.1)	< .001	0.2 (0.00–5.3)	0.2 (0.0–3.4)	.264
Age group
Children	0.6 (0.0–1.8)	0.6 (0.0–1.6)	.741	0.4 (0.0–2.0)	0.4 (0.0–1.1)	.071
Adolescent	1.8 (0.5–4.7)	2.3 (0.7–6.0)	.003	1.9 (0.5–5.9)	2.0 (0.7–5.9)	.332
Adult	85.4 (78.0–90.8)	82.8 (75.0–88.7)	< .001	76.5 (67.0–84.2)	78.4 (70.8–85.1)	.006
Older adult	8.5 (4.2–13.9)	11.2 (6.6–16.9)	< .001	15.8 (8.6–23.3)	15.8(10–22.7)	.343
Opioid Prescribing,§ Median (Q1-Q3)
Patient-based opioid prescribing rate	95.5 (87.8–104.2)	91.6 (80.9–101.3)	< .001	43.3 (28.2–58.1)	81.8 (69.5–92.3)	< .001
High-risk prescribing, %
Morphine milligram equivalent > 50/d	8.1 (2.0–26.5)	10.2 (2.4–30.5)	.020	2.1 (0–8.5)	3.5 (0.8–11.4)	< .001
Morphine milligram equivalent > 90/d	0.0 (0.0–1.3)	0.4 (0.0–1.8)	.001	0.00 (0.0–0.1)	0.0 (0.0–0.8)	< .001
> 3 d	21.8 (8.9–51.7)	26.5 (10.3–56.8)	.032	9.01 (2.2–26.2)	15.3 (4.9–45.2)	< .001
> 7 d	0.5 (0.0–1.7)	0.5 (0.0–1.5)	.891	0.00 (0.0–0.0)	0.0 (0.0–0.5)	< .001

^*Total N = 6,845; decliners, n = 513; all other groups, n = 6,332.

^†Because medians and 25th and 75th percentiles (Q1-Q3) are presented, the totals may not add to 100%.

^‡P values derived from Mann-Whitney U test.

^§Patient mix and opioid prescribing measures were calculated at the dentist level and then aggregated to the group level.

^¶25th percentile-75th percentile (Q1-Q3).

^#Includes all prescriptions (opioids and nonopioids).

DISCUSSION

In this study, we aimed to identify dentists’ opioid prescribing trajectories and characterize high opioid prescribers using a longitudinal national prescribing data set from 2015 through 2019 and GBTM. As in other studies, we observed an overall decreasing trend of opioid prescribing.^13,14 However, with our GBTM analysis, we were able to define distinct groups of prescribers and show differences in the rates of decline among the trajectory groups over 5 years.

As we hypothesized, GBTM identified a group of dentists who rarely engaged in opioid prescribing (< 1% of opioids) over the study, a small group of dentists (3.4%) who had the highest opioid prescribing rates at baseline and who continued to prescribe at high rates, and a group of dentists who decreased their prescribing over time. Chua and colleagues²⁸ similarly found a small percentage of dentists (5%) to be high-volume opioid prescribers in 2019. However, Chua and colleagues’ analysis used an arbitrary cutoff point of prescribers in the 95th percentile to define high prescribers and used only 1 year of data (2019), whereas we used GBTM to define trajectories with 5 years of longitudinal data. Like Chua and colleagues’ findings,²⁸ not only were these dentists high-volume opioid prescribers, but a greater proportion of their opioid prescriptions also were considered high risk (≥50 MME per day). This is of concern because taking an opioid that is prescribed at 50 or more MME per day doubles the risk of overdose-related death, per the CDC.

We observed that OMFSs were concentrated in the moderately and consistently high prescribing groups. Our finding may reflect the nature of procedures performed by OMFSs, which are more invasive and thus are perceived to be associated with higher pain levels. It is also possible that patient populations served by OMFSs have higher rates of NSAID contraindications. Due to data limitations, we were unable to examine these factors. Given the strong evidence supporting NSAIDs based on the impacted third molar model,^5,29,30 there is an opportunity to improve prescribing for dental pain among OMFSs. Although a greater proportion of OMFSs were consistently high prescribers, 81% of the dentists in the group were general dentists. Thus, interventions may need to be tailored to high-prescribing general dentists as well.

Although previous studies found that overall opioid prescribing by dentists and OMFSs has decreased by 41% and 20%, respectively, from 2016 through 2019,^13,14 we found that those rates of decrease were not distributed evenly across dentists. With the exception of 1 group of rapid decliners, the higher a group’s baseline rate of prescribing, the less the members of that group decreased their prescribing over time. As described above, the higher a group’s baseline rate of prescribing, the more the group was composed of OMFSs. Thus, our findings are consistent with studies that found that opioid prescribing rates by OMFSs decreased at lower rates than by general dentists and that the rate of opioid prescribing was highest for surgical dental procedures.^14,31,32 By the end of our study, the opioid prescribing rate of the high-prescribing group remained higher than the baseline opioid prescribing rate of any of the other groups.

Within the consistently high prescribers, a small number (n = 513) decreased their rate of opioid prescribing by almost 60%. Although the combined other groups decreased their prescribing rates only minimally, by the end of the study, their high-risk prescribing was lower than the rapid decliners at the start of the study. If the combined other groups’ change in the median rate of high-risk prescribing indicates receptiveness to changing their opioid prescribing, then interventions may be effective. Understanding who the rapid decliners were may provide insights on intervention points to target those in the combined other groups. Within the variables available in our data, we found that the groups were mostly similar at baseline. However, rapid decliners at baseline had a higher proportion of Medicaid reimbursements, which may suggest that Medicaid’s opioid prescribing policies reduced opioid prescribing in this group.^33–35 Thus, policy changes may be effective. In addition to unmeasured confounding variables, other factors that were not available in our analysis and mentioned in the literature may explain differences between groups, such as provider training or year of graduation, provider sex, state policies, and patients’ medical history.^36–38

Barriers to decrease opioid prescribing among dentists have been reported to include managing the treatment of patients expecting to receive opioids, fear of loss of business, treating patients with comorbidities, just-in-case prescribing, and managing pain in patients with substance use disorders.^39,40 Strategies that have been successful in curbing opioid prescribing in dentistry include CE programs, PDMPs, and default tablet dispensing quantities in electronic health records.^41–43 However, not all state dental boards require CE on opioids for licensure and relicensure, there is variation in PDMP use; and there is no consensus in the recommended quantity of opioids for dental pain.⁴⁴ Moreover, there have been no national guidelines for dental pain prescribing so far. These factors may explain why some dentists have not changed their opioid prescribing behavior. The upcoming American Dental Association guidelines for dental pain management could play an important role in limiting the use of opioids to when they are necessary and may accelerate the cultural change needed to make NSAIDs the norm for dental pain management.

Due to data limitations, we could not determine the procedures associated with opioid prescriptions and were unable to adjust for comorbidities such as chronic pain (ie, temporomandibular joint disorders) or NSAID contraindications. We did not adjust for policy effects such as CDC guidelines, state laws, or licensure requirements. Future studies should evaluate whether policies can change dentist behavior. Furthermore, our data do not capture the period of the COVID-19 pandemic (2020–2023). However, we believe including opioid prescribing for dental pain during the pandemic might present a skewed picture of opioid prescribing due to disruptions in access to care.

CONCLUSIONS

Using national prescribing data from 2015 through 2019, we identified several opioid-prescribing trajectories confirming that variations exist in dentists’ opioid prescribing. GBTM identified a group of consistently high opioid prescribers and a small number who rapidly decreased prescribing. Compared with the nondecliners group at baseline, the decliners group included more Medicaid prescription reimbursement and fewer OMFSs. However, decliners and nondecliners were otherwise similar, suggesting unmeasured factors may contribute to differences in prescribing behavior. Further understanding of high opioid prescribers’ characteristics is needed to create targeted interventions.

ABBREVIATION KEY

CDC

Centers for Disease Control and Prevention

CE

Continuing education

GBTM

Group-based trajectory modeling

LRx

IQVIA Longitudinal Prescription Data

MME

Morphine milligram equivalents

NSAID

Nonsteroidal anti-inflammatory drug

OMFS

Oral and maxillofacial surgeon

PDMP

Prescriptions drug monitoring program

Q1-Q3

25th percentile-75th percentile.

Biographies

Dr. Khouja was a postdoctoral associate, Division of General Internal Medicine, School of Medicine, University of Pittsburgh, Pittsburgh, PA, when the work described in this article was conducted. She now is a clinical assistant professor, Department of Community Dentistry and Behavioral Science, University of Florida College of Dentistry, Gainesville, FL.

Dr. Shah is an assistant professor, Department of Dental Public Health, University of Pittsburgh School of Dental Medicine, Pittsburgh, PA.

Dr. Suda is a professor, Division of General Internal Medicine, University of Pittsburgh School of Medicine, Pittsburgh, PA, and the associate director, Center for Health Equity Research and Promotion, VA Pittsburgh Healthcare System, Pittsburgh, PA.

Dr. Polk is an assistant professor, Department of Dental Public Health, School of Dental Medicine, University of Pittsburgh, Pittsburgh, PA. Address correspondence to Dr. Polk, Department of Dental Public Health, School of Dental Medicine, University of Pittsburgh, 3501 Terrace St, 381 Salk Hall, Pittsburgh, PA 15261.

APPENDIX. GROUP-BASED TRAJECTORY MODELING MODEL SELECTION

To select the best model that represents trajectories of opioid prescribing among all dentists in our cohort (model 1), we tested group-based trajectory modeling (GBTM) from 2 through 12 groups and opioid prescription rates as a smooth function of time. For trajectories within the consistently high prescribing group (model 2), we tested GBTM from 2 through 5 groups. We determined the degree of the fitted polynomial by means of examining all possible permutations up to fourth order polynomials. Because our outcome of opioid prescribing rates was a continuous variable with excess zeros, we used a 0 inflated Poisson to analyze group membership. We determined the number of groups and degree of polynomial in each trajectory group using the Bayesian Information Criterion (BIC) to measure improvement in model fit. We chose the best model using comparisons of estimated and observed prescribing trajectories, BIC, an average posterior probability greater than 0.7, and clinical relevance.

For model 1, we ran linear models with the number of groups ranging from 2 through 12. Models with higher order polynomials did not converge. BICs and other model-fit criteria did not indicate a best model (eTable 1). Therefore, we conducted the following steps to select the best model. As we hypothesized, despite the number of groups included, GBTM identified a consistently low and a consistently high prescribing group. We compared the percentage of providers in the lowest and highest trajectories within the 2- through 12-group models. We observed similar percentages of high and low prescribers within the 7-, 8-, and 9-group models. Starting with the 8-group model, a new pattern of change over time appeared that was not present with the 7-group model. In the 9-group model, that new pattern split into 2 trajectories with similar slopes (eFigure 2). High confidence in the accuracy of group membership classification was achieved on the basis of the maximum posterior probability assignment rule (eTable 2).

For model 2, we ran linear models from 2 through 5 groups for the 6,845 consistently high prescribers identified in the previous model. GBTM did not converge with higher order polynomials or beyond 4 groups. On the basis of the model fit criteria, we selected the 4-group model (eTable 4). High confidence in the accuracy of group membership classification was achieved on the basis of the maximum posterior probability assignment rule (eTable 5).

eFigure 1.

Flowchart of cohort selection.

eFigure 2.

Group-based trajectory modeling, model 1 selection. A. Two-group model. B. Three-group model. C. Four-group model. D. Five-group model. E. Six-group model. F. Seven-group model. G. Eight-group model. H. Nine-group model. I. Ten-group model. J. Eleven-group model. K. Twelve-group model.

eTable 1.

Bayesian Information Criterion values for model 1 selection.

MODEL	BAYESIAN INFORMATION CRITERION
2 groups	−3,699,261
3 groups	−3,198,197
4 groups	−2,984,571
5 groups	−2,867,283
6 groups	−2,807,578
7 groups	−2,773,575
8 groups	−2,740,398
9 groups	−2,718,573
10 groups	−2,700,808
11 groups	−2,687,867
12 groups	−2,678,044

eTable 2.

Average posterior probability of assignment by group based on maximum posterior probability rule, model 1 group-based trajectory modeling (N = 199,145).

GROUP	AVERAGE POSTERIOR PROBABILITY (% OF TOTAL)
	Nonprescribers (14.8%)	Low Prescribers 1 (15.2%)	Low Prescribers 2 (17.7%)	Low Prescribers 3 (13.1%)	Decliners (7.3%)	Moderately High Prescribers 1 (16.4%)	Moderately High Prescribers 2 (12.1%)	Consistently High Prescribers (3.4%)
Nonprescribers	0.92
Low Prescribers 1		0.91
Low Prescribers 2			0.89
Low Prescribers 3				0.86
Decliners					0.90
Moderately High Prescribers 1						0.87
Moderately High Prescribers 2							0.92
Consistently High Prescribers								0.95

eTable 3.

Group-based trajectory modeling model 1 estimates and change in opioid prescribing rates in 2019 compared with rates in 2015 (N = 199,145).^*

GROUP (% OF TOTAL)	YEAR	OBSERVED	PREDICTED	95% CI	CHANGE (%), 2019 VS 2015
Nonprescribers (14.8)	2015	0.90	0.89	0.87 to 0.90
	2016	0.54	0.55	0.54 to 0.56
	2017	0.33	0.34	0.33 to 0.35
	2018	0.20	0.21	0.21 to 0.22
	2019	0.15	0.13	0.13 to 0.13	−83.80
Low Prescribers 1 (15.2)	2015	5.5	5.7	5.6 5.7
	2016	4.1	4.1	4.1 to 4.1
	2017	2.9	3.0	2.9 to 3.0
	2018	2.0	2.1	2.1 to 2.1
	2019	1.5	1.5	1.5 to 1.5	−72.93
Low Prescribers 2 (17.7)	2015	12.3	12.7	12.6 to 12.8
	2016	10.2	9.9	9.9 to 10.0
	2017	7.9	7.8	7.7 to 7.8
	2018	5.7	6.0	6.0 to 6.1
	2019	4.4	4.7	4.7 to 4.7	−64.44
Low Prescribers 3 (13.1)	2015	16.9	17.4	17.4 to 17.5
	2016	16.4	16.1	16.0 to 16.1
	2017	15.5	14.7	14.7 to 14.8
	2018	13.7	13.5	13.4 to 13.6
	2019	11.9	12.3	12.2 to 12.4	−29.77
Decliners (7.3)	2015	29.4	32.2	32.0 to 32.3
	2016	24.2	21.5	21.4 to 21.6
	2017	16.4	14.3	14.3 to 14.4
	2018	8.7	9.5	9.5 to 9.6
	2019	5.1	6.3	6.3 to 6.4	−82.58
Moderately High Prescribers 1 (16.4)	2015	28.7	30.0	29.9 to 30.1
	2016	27.9	26.7	26.6 to 26.8
	2017	25.7	23.8	23.7 to 23.8
	2018	21.8	21.1	21.0 to 21.2
	2019	18.1	18.6	18.6 to 18.7	−36.86
Moderately High Prescribers 2 (12.1)	2015	39.2	40.4	40.3 to 40.5
	2016	38.7	37.3	37.3 to 37.4
	2017	36.7	34.4	34.3 to 34.5
	2018	32.8	31.6	31.5 to 31.7
	2019	28.8	28.9	28.8 to 29.0	−26.60
Consistently High Prescribers (3.4)	2015	54.6	55.8	55.6 to 56.0
	2016	54.6	53.0	52.8 to 53.1
	2017	52.9	50.2	50.0 to 50.3
	2018	49.2	47.4	47.2 to 47.6
	2019	44.7	44.6	44.4 to 44.8	−18.05

^*Model estimates are presented as rates and were derived from a group-based trajectory modeling zero inflated Poisson linear model.

eTable 4.

Bayesian Information Criterion values for model 2 selection.

MODEL	BAYESIAN INFORMATION CRITERION
2 Groups	−107,730.6
3 Groups	−106,461.3
4 Groups	−105,278.4

eTable 5.

Average posterior probability of assignment by group based on maximum posterior probability rule, model 2 group-based trajectory modeling (n = 6,845).

GROUP (% OF TOTAL)	AVERAGE POSTERIOR PROBABILITY
	Rapid Decliners	Other Group 1	Other Group 2	Other Group 3
Rapid Decliners (7.5)	0.84
Other Group 1 (20.9)		0.84
Other Group 2 (5.0)			0.93
Other Group 3 (66.7)				0.91

eTable 6.

Group-based trajectory modeling model 2 estimates and change in opioid prescribing rates in 2019 compared with 2015 (n = 6,845).

YEAR	OTHER GROUP 1 (20.9% OF TOTAL)			RAPID DECLINERS (7.5% OF TOTAL)			OTHER GROUP 2 (5.0% OF TOTAL)			OTHER GROUP 3 (66.7% OF TOTAL)
	Observed	Predicted	95% CI	Observed	Predicted	95% CI	Observed	Predicted	95% CI	Observed	Predicted	95% CI
2015	61.0	62.6	62.1 to 63.1	66.0	72.40	71.7 to 73.1	50.1	51.3	51.1 to 51.5	77.3	79.0	78.0 to 80.1
2016	61.6	60.8	60.4 to 61.2	62.7	59.28	58.8 to 59.8	50.0	49.4	49.3 to 49.6	79.0	77.8	77.0 to 78.6
2017	60.6	59.0	58.7 to 59.4	54.1	48.51	48.1 to 48.9	49.0	47.6	47.5 to 47.8	78.0	76.5	75.8 to 77.2
2018	57.8	57.2	56.8 to 57.6	40.6	39.66	39.3 to 40.0	46.3	45.8	45.7 to 46.0	75.7	75.2	74.3 to 76.0
2019	53.9	55.4	55.0 to 55.9	26.5	32.38	32.1 to 32.7	43.1	44.0	43.9 to 44.2	71.1	73.8	72.7 to 74.9
Change 2019 vs 2015, %	−11.7	NA*	NA	−59.9	NA	NA	−14.0	NA	NA	−8.0	NA	NA

^*NA: Not applicable.