Comparative Study of Opioid Initiation With Tramadol, Short-acting Hydrocodone, or Short-acting Oxycodone on Opioid-related Adverse Outcomes Among Chronic Noncancer Pain Patients

Abstract

Objective: 

To compare the safety profiles of low and high-dose tramadol, short-acting hydrocodone, and short-acting oxycodone therapies among chronic noncancer pain individuals.

Materials and Methods: 

A retrospective cohort study of individuals with back/neck pain/osteoarthritis with an initial opioid prescription for tramadol, hydrocodone, or oxycodone was conducted using IQVIA PharMetrics Plus claims for Academics database (2006 to 2020). Two cohorts were created for separately studying opioid-related adverse events (overdoses, accidents, self-inflicted injuries, and violence-related injuries) and substance use disorders (opioid and nonopioid). Patients were followed from the index date until an outcome event, end of enrollment, or data end. Time-varying exposure groups were constructed and Cox regression models were estimated.

Results: 

A total of 1,062,167 (tramadol [16.5%], hydrocodone [61.1%], and oxycodone [22.4%]) and 986,809 (tramadol [16.5%], hydrocodone [61.3%], and oxycodone [22.2%]) individuals were in the adverse event and substance use disorder cohorts. All high-dose groups had elevated risk of nearly all outcomes, compared with low-dose hydrocodone. Compared with low-dose hydrocodone, low-dose oxycodone was associated with a higher risk of opioid overdose (hazard ratio: 1.79 [1.37 to 2.33]). No difference in risk was observed between low-dose tramadol and low-dose hydrocodone (hazard ratio: 0.85 [0.64 to 1.13]). Low-dose oxycodone had higher risks of an opioid use disorder, and low-dose tramadol had a lower risk of accidents, self-inflicted injuries, and opioid use disorder compared with low-dose hydrocodone.

Discussion: 

Low-dose oxycodone had a higher risk of opioid-related adverse outcomes compared with low-dose tramadol and hydrocodone. This should be interpreted in conjunction with the benefits of pain control and functioning associated with oxycodone use in future research.

Opioid prescribing for chronic noncancer pain increased from 13% in 2001 to 23% per chronic pain visit in 2010.¹ However, a linear decline in opioid prescribing was observed until 2016, after which the rate of decline in opioid prescribing increased, in part due to the publishing of the Centers for Disease Control and Prevention (CDC) opioid prescribing guideline in 2016.² The recent decrease in opioid prescribing is shaped by fewer initiations of opioid prescriptions.³ Initiation of prescription opioids decreased from 1.6% in 2012 to 0.8% in 2017 among all enrollees of a large commercial health plan.³ With this downward trend, the focus has shifted towards balancing the benefits of pain relief and better functioning with the risk of overdose, opioid use disorder, and opioid-driven accidents and injuries.^4,5 One key factor in the opioid benefit-risk equation could be the type of initial opioid prescribed.

In 2014, hydrocodone (57%) was the most frequently prescribed opioid on the first pain-related visit, followed by tramadol (32%) and oxycodone (10%) based on commercial insurance claims data.⁶ These 3 drugs differ in their pharmacokinetic and pharmacodynamic properties.⁷ Tramadol has a dual mechanism of action, with opioid agonist activity and serotonin-norepinephrine reuptake inhibition properties.⁸ Desmethyltramadol, an active metabolite of tramadol, has strong opioid mu activity, which makes the efficacy and safety of tramadol dependent on cytochrome P450 metabolism.⁸ Hydrocodone has an agonist activity on opioid mu receptors at the typically prescribed dose, while also having activity on opioid delta and kappa receptors at high doses.⁹ Oxycodone also primarily activates opioid mu receptors.¹⁰ In the United States, oxycodone and hydrocodone are schedule II drugs, whereas tramadol is a schedule IV drug, highlighting the belief that tramadol is considered to confer a lower risk of abuse compared with hydrocodone and oxycodone.^11,12

A few self-controlled studies have evaluated the abuse potential of tramadol, hydrocodone, and oxycodone.^13,14 Healthy adult volunteers with documented opioid abuse have been the participants in these studies, wherein physiological and self-reported outcomes were reported.^13,14 The studies reported similar abuse liability of each of the 3 opioid analgesics.^13,14 However, it has been shown that abuse potential could be dose-dependent, as higher doses of oxycodone induced more abuse-type of symptoms, quantified using physiological measures such as respiratory rates and pupil size, as compared with low-dose oxycodone and hydrocodone.¹³ In addition, a randomized trial comparing tramadol and hydrocodone on abuse index (defined as dose escalation without physician indication, loss of control, opioid withdrawal, or nonmedical use) reported that 2.7% and 4.9% of patients randomized to tramadol and hydrocodone, respectively had an abuse score (P < 0.01).¹⁵ However, in that study, physicians could switch to or add other opioids after the first randomized opioid, which could have biased the comparison, especially if the comparison of interest is tramadol and hydrocodone-only therapy over the duration of follow-up. As a result, clinical studies assessing the safety profile of commonly prescribed opioids based on the number of days on each opioid therapy and the dose prescribed are warranted.

The objective of this study was to compare the risks of opioid-related adverse events (AE) (overdoses, accidents, self-inflicted injuries, and violence-related injuries) and substance use disorder outcomes (opioid use disorder and nonopioid substance use disorder), among patients initiating tramadol, short-acting hydrocodone, and short-acting oxycodone therapies, accounting for the prescribed dose and the number of days in each group.

MATERIALS AND METHODS

Data

This study used a 10% random sample of the IQVIA PharMetrics Plus for Academics database (2006 to 2020), an administrative health insurance claims database containing information on enrollees in commercial, Medicare Advantage, and Medicaid managed care plans in the United States. Given the deidentified nature of the data used, the study was considered not to be human subjects research.

Study Design and Cohort Selection

A retrospective cohort study of adults with chronic noncancer pain (back pain, neck pain, or osteoarthritis) who initiated opioid therapy on tramadol, short-acting hydrocodone, or short-acting oxycodone with or without being combined with acetaminophen was conducted. First, individuals with an initial opioid prescription for tramadol, short-acting hydrocodone, or short-acting oxycodone between July 01, 2006 and May 30, 2020 were identified. Opioid prescriptions were identified using Generic Product Identifier codes.¹⁶ The type of formulation (short-acting/long-acting) was identified using the dosage form variable. The date of the first opioid prescription was considered the index date. Individuals were required to be continuously enrolled with medical and pharmacy benefits for 6 months preceding the index date (baseline period). As the target population of the study was patients with chronic back/neck pain or osteoarthritis, patients were required to have 2 diagnosis codes for the same condition of back pain, neck pain, or osteoarthritis. The first diagnosis was required 90 days before or on the index date, and the second diagnosis was required within 30 to 180 days of the first diagnosis in either baseline or follow-up period. These pain conditions were identified in the inpatient or outpatient setting using International Classification of Diseases (ICD)-9-CM or ICD-10-CM diagnosis codes (Supplement Table 1, Supplemental Digital Content 1, https://links.lww.com/CJP/A901). Individuals with missing dates of birth or missing sex were excluded. Also, the study sample was restricted to individuals 18 years or older to focus on the adult population. To limit the analyses to individuals without cancer and certain pain conditions with complex therapeutic needs, individuals with the following conditions or health service utilization in the baseline were excluded: (1) cancer, (2) hospice care, (3) long-term care, (4) rheumatoid arthritis, (5) spinal injuries, (6) pregnancy, or (7) organ transplant. Individuals with opioid prescriptions with negative or >180 days’ supply or >1000 units were excluded from the sample, as these prescriptions are likely to be erroneous. Individuals with more than one opioid drug dispensed on the index date (eg, tramadol plus morphine or hydrocodone plus oxycodone) were excluded.

Two cohorts were then created for separately studying opioid-related AEs (overdose, accidents, self-inflicted injuries and poisoning, and violence-related injuries) and substance use disorder outcomes (opioid use disorder and nonopioid substance use disorder). To identify incident cases for our outcome measures, individuals with opioid overdose in the baseline were excluded from the AEs cohort. For the substance use disorder cohort, individuals with substance use disorder in the baseline (opioid or nonopioid) were excluded. This was done to retain individuals with prior substance use disorder diagnoses to study opioid-related AEs and to retain individuals with prior overdoses to study substance use disorder outcomes. Individuals in both cohorts were followed until the first incidence of the outcome, enrollment end date, or data end date, whichever was the earliest.

Study Measures

Exposure

Low and high-dose tramadol, short-acting hydrocodone, and short-acting oxycodone were the main exposures of interest. A time-varying approach was used to construct time windows of exposure to 13 dose and drug-based categories. First, 7 exposure groups were created, based on which opioid was received on each day in the follow-up period: (1) tramadol only, (2) short-acting hydrocodone only, (3) short-acting oxycodone only, (4) tramadol combination, (5) nontramadol opioid combination, (6) other opioids, and (7) no opioid. The other opioids group primarily included codeine, fentanyl, morphine, methadone, tapentadol, hydromorphone, oxymorphone, and propoxyphene, and long-acting formulations of oxycodone and hydrocodone. Each of the 6 opioid therapy groups were then divided into low-dose and high-dose based on the average morphine milligrams equivalent (MME) per day during the respective time windows resulting in 13 exposure groups. An average daily dose of <50 MME was considered a low dose and ≥50 MME was considered a high dose.¹⁷ To better attribute transitions from one opioid to another or a different dose to the outcomes, the treatment windows were adjusted using 30% of days’ received by each opioid exposure group for a maximum of 30 days. A previous study used a fixed 14-day period to extend the total days on opioids, based on the finding that around 40% of the AEs occurred in that period.¹⁸ We used the percentage of days’ supply for adjusting the treatment period, instead of a fixed number of days, based on the clinical judgment that it better reflects the actual relationship between opioid therapy and the adverse outcomes. In our approach, for example, if a person had low-dose short-acting oxycodone prescriptions for 120 days and no opioid treatment for 80 days starting from the 121st day, the time was adjusted as followed: 120 + 30 days of low-dose short-acting oxycodone (0.3×120=36, which is >30, so 30 days is used for adjustment) and 50 days of no opioid treatment.

Outcomes

For the AEs cohort, the outcomes studied were opioid overdose, accidents, self-inflicted injuries and poisoning, and violence-related injuries. In the primary analysis of overdose, overdose was identified using published ICD-9-CM and ICD-10-CM diagnosis codes (main overdose definition).¹⁹ Validation studies of these codes, which used physician charts as the gold standard, reported the sensitivity, specificity, and positive predictive values as 25%, 99%, and 81%, respectively.^20,21 Because of the low sensitivity of the overdose diagnosis codes, 2 additional broader measures of opioid overdose were used in sensitivity analyses. In the first broad measure, procedure codes for naloxone administration in the emergency department and diagnosis codes for opioid-adverse effects in any setting were used, in addition to the overdose diagnosis codes. In the second, and the broadest overdose measure, diagnosis for respiratory adverse effects (any setting), mechanical ventilation procedures (emergency department only), or diagnosis for central nervous system adverse effects (any setting) for individuals younger than 50 years of age were considered in addition to the first broad measure.

Because misuse of prescription opioids could manifest in events such as accidents, violence-related injuries, and self-inflicted injuries, these outcomes were evaluated as well. Accidents and violence-related injuries were identified using ICD-9-CM and ICD-10-CM codes obtained from a published study.¹⁹ Self-inflicted injuries and poisoning outcomes were identified using a published algorithm, which uses a combination of suicide and poisoning codes in addition to hospitalization information.²²

For the substance use disorder cohort, the outcomes were opioid use disorder and other substance use disorders. Opioid use disorder was identified using ICD-9-CM and ICD-10-CM diagnosis codes, based on existing literature.^19,23 The other substance use disorder outcome included use disorders of benzodiazepines, amphetamines, marijuana, cocaine, alcohol, and other psychoactive drugs.^19,23 All diagnosis and procedure codes used for identifying the outcomes are provided in Supplement Table 1 (Supplemental Digital Content 1, https://links.lww.com/CJP/A901).

Covariates

Age and sex have consistently been associated with opioid-related adverse outcomes.²⁴ Substantial geographic variation exists in opioid overdose, opioid use disorder, and other substances use disorders.²⁵ As a result, patient age, sex, and region of residence were used as covariates. Opioid use disorder was reported to be higher in unemployed and individuals with lower incomes.²⁶ As a proxy for economic status, a payer-type variable was used, which included coverage by Medicaid. To account for the fact that sicker individuals have a higher risk of opioid-related events, the Charlson Comorbidity Index was used.²⁷ Benzodiazepines,²⁸ muscle relaxants,²⁹ hypnotics,³⁰ antidepressants,³¹ and gabapentinoids³² have been shown to increase the risk of opioid-related adverse outcomes and were used in the adjusted analyses. Mental health disorders are strongly correlated with substance use disorders and opioid overdose.³³ The following mental health disorders were included as covariates: (1) anxiety disorders, (2) personality disorders, (3) mood disorders, (4) major depressive disorder, (5) posttraumatic stress disorder, (6) nicotine dependence, and (7) schizophrenia. (Supplement Table 1, Supplemental Digital Content 1, https://links.lww.com/CJP/A901). All covariates were gathered from the baseline period.

State-level fatal drug overdose rates were also adjusted for in the models. Fatal drug overdose rates corresponding to the state of residence and the index year of the individuals in our sample were identified using the CDC Wonder database.^34,35

Sensitivity Analyses

We conducted sensitivity analyses to test the robustness of the findings to changes in exposure definition and statistical approach. In the first sensitivity analysis, time-varying exposure groups were constructed using the actual days in each exposure group without extending the exposure windows (the 30% approach) as performed in the main analysis. Second, an intent-to-treat type of analysis was conducted using the type of initial opioid as the exposure group (3 groups: tramadol only, short-acting hydrocodone only, and short-acting oxycodone only), regardless of dose. The individuals were followed until the run-out date of any opioid prescription + 180 days, the incidence of the outcome, loss of enrollment, or end of data, whichever was the earliest. The 180-day additional follow-up window was selected, as previous studies have shown that patients are at elevated risk for opioid-related AEs after opioid discontinuation.^36,37 A study by Oliva et al³⁷ showed that the risk of overdose decreased monotonically until 180 days after opioid discontinuation and remained stable in the subsequent period. Third, the intent-to-treat analysis was conducted using propensity score adjustment. All previously described covariates were used to construct the propensity score for receiving hydrocodone using multinomial logistic regression, and this propensity score was used as the only covariate in the final analysis for the comparison across the three groups. Fourth, a falsification test was conducted to assess the presence and extent of unmeasured confounding. The outcome of dermatitis was used as a falsification outcome because of its implausible association with the use of individual opioid drugs.

Statistical Analyses

Time-varying Cox regression models were estimated using the previously described treatment windows as the main variables of interest. Separate analyses were conducted for each outcome using the respective cohorts. Both unadjusted and fully adjusted multivariable analyses were performed, with the multivariable models adjusting for demographics, medication use, pain conditions, and mental health conditions previously described. To fully portray the risks of the opioid exposure groups, separate Cox regressions were performed using 2 reference groups: no opioid and low-dose short-acting hydrocodone only. Hazard ratios (HR) and 95% CI are reported. All analyses were conducted using SAS v9.4.

RESULTS

A total of 3,972,126 individuals had an initial opioid prescription for tramadol, short-acting hydrocodone, or short-acting oxycodone with no opioid prescriptions in the 6 months prior. After applying the common inclusion and exclusion criteria for both cohorts, 1,063,887 individuals remained in the overall sample (Table 1). With the requirement of no opioid overdose in the baseline period, 1,062,167 individuals were included in the AEs cohort. A total of 986,809 individuals were included in the substance use disorder cohort. Approximately 61% initiated opioid use on short-acting hydrocodone, 22% on short-acting oxycodone, and 17% on tramadol in both cohorts.

TABLE 1 - Flow Diagram for Sample Selection for the 2 Cohorts

Individuals with initial opioid prescription for short-acting hydrocodone, short-acting oxycodone, or tramadol	5,328,474
Individuals with 6 mo of enrollment before the initial opioid prescription (index date)	3,972,126
Individuals with back pain, neck pain, or osteoarthritis in 90 d before and including index date	1,490,094
Individuals with second diagnosis for the same condition of back pain, neck pain, or osteoarthritis within 30-180 d of first diagnosis	1,430,703
Individuals with age at index date ≥18 y	1,276,074
Individuals with nonmissing sex	1,275,984
Individuals without cancer in the 6 mo pre period (baseline)	1,218,843
Individuals without hospice claims in baseline	1,218,632
Individuals without long-term care claims in baseline	1,218,152
Individuals without rheumatoid arthritis in baseline	1,198,356
Individuals without paraplegia/quadriplegia in baseline	1,193,799
Individuals without pregnancy in baseline	1,168,293
Individuals without organ transplant in baseline	1,165,859
Individuals without potentially erroneous opioid prescriptions in the entire follow-up (negative or >180 days’ supply; negative or >1000 quantities)	1,078,020
Individuals with only tramadol, short-acting hydrocodone, or short-acting oxycodone on the index date	1,063,887
Opioid-related adverse events cohort
Individuals without opioid overdose (the broadest definition) in the baseline	1,062,167
Initiated on short-acting hydrocodone; n (%)	648,902 (61.1)
Initiated on short-acting oxycodone; n (%)	237,762 (22.4)
Initiated on tramadol; n (%)	175,503 (16.5)
Substance use disorder cohort
Individuals without opioid use disorder in the baseline	1,058,842
Individuals without other substance use disorder in the baseline	986,809
Initiated on short-acting hydrocodone; n (%)	604,991 (61.3)
Initiated on short-acting oxycodone; n (%)	218,624 (22.2)
Initiated on tramadol; n (%)	163,194 (16.5)

Approximately 63% of the tramadol initiators in both cohorts were female, although ~54% of hydrocodone and oxycodone initiators were female (Table 2). More than 70% had chronic osteoarthritis in all 3 initiator groups whereas back pain was observed more frequently in tramadol initiators (around 48%) compared with hydrocodone (40%) and oxycodone (36%) initiators. Around 28% of tramadol initiators in either cohort had skeletal muscle relaxant prescriptions, whereas 21% and 16% of hydrocodone and oxycodone initiators used had skeletal muscle relaxants in the baseline. The total opioid prescriptions and the type of formulations of tramadol, short-acting hydrocodone, and short-acting oxycodone are reported in Supplement Table 2 (Supplemental Digital Content 2, https://links.lww.com/CJP/A902).

TABLE 2 - Baseline Characteristics of the Opioid-related Adverse Events (n = 1,062,167) and Substance Use Disorder Cohorts (n = 986,809)

	Cohort for assessing opioid-related adverse events						Cohort for assessing substance use disorder outcomes
	Tramadol (n=175,503)		Hydrocodone (n=648,902)		Oxycodone (n=237,762)		Tramadol (n=163,194)		Hydrocodone (n=604,991)		Oxycodone (n=218,624)
Characteristics	n	Percent	n	Percent	n	Percent	n	Percent	n	Percent	n	Percent
Age (y)
Mean (SD)	49.44 (14.2)		47.42 (14.2)		48.18 (14.1)		49.71 (14.2)		47.59 (14.3)		48.35 (14.2)
18-30	21,203	12.1	97,582	15.0	32,887	13.8	19,146	11.7	89,654	14.8	29,943	13.7
31-44	39,851	22.7	158,829	24.5	55,865	23.5	36,556	22.4	146,482	24.2	50,756	23.2
45-54	44,055	25.1	165,661	25.5	60,837	25.6	40,752	25.0	153,888	25.4	55,493	25.4
55-64	45,899	26.2	157,201	24.2	61,075	25.7	43,182	26.5	148,239	24.5	56,730	26.0
65 and above	24,495	14.0	69,629	10.7	27,098	11.4	23,558	14.4	66,728	11.0	25,702	11.8
Sex
Female	110,055	62.7	351,499	54.2	123,625	52.0	103,325	63.3	330,187	54.6	114,953	52.6
Male	65,448	37.3	297,403	45.8	114,137	48.0	59,869	36.7	274,804	45.4	103,671	47.4
Region of residence
East	33,654	19.2	104,791	16.2	72,872	30.7	31,465	19.3	98,027	16.2	67,981	31.1
Mid West	51,847	29.5	195,299	30.1	62,777	26.4	47,365	29.0	178,889	29.6	56,171	25.7
South	54,611	31.1	165,737	25.5	53,226	22.4	51,474	31.5	155,932	25.8	49,448	22.6
West	30,055	17.1	161,052	24.8	46,696	19.6	28,066	17.2	152,158	25.2	43,036	19.7
Missing	5336	3.0	22,023	3.4	2191	0.9	4824	3.0	19,985	3.3	1988	0.9
Insurance type
Commercial only	125,703	71.6	513,566	79.1	188,134	79.1	119,121	73.0	484,875	80.2	175,346	80.2
Medicaid only	19,679	11.2	44,931	6.9	13,960	5.9	15,624	9.6	34,929	5.8	9966	4.6
Medicare only	15,307	8.7	40,101	6.2	15,855	6.7	14,299	8.8	37,220	6.2	14,534	6.7
Others/missing	1609	0.9	6157	1.0	2423	1.0	1486	0.9	5725	1.0	2239	1.0
Self-insured only	13,205	7.5	44,147	6.8	17,390	7.3	12,664	7.8	42,242	7.0	16,539	7.6
Index year
2006	6476	3.7	37,834	5.8	11,715	4.9	6251	3.8	36,057	6.0	11,132	5.1
2007	15,032	8.6	84,129	13.0	26,527	11.2	14,345	8.8	79,872	13.2	24,930	11.4
2008	15,706	9.0	83,448	12.9	27,226	11.5	14,910	9.1	79,070	13.1	25,469	11.7
2009	14,560	8.3	73,779	11.4	25,515	10.7	13,832	8.5	69,756	11.5	23,946	11.0
2010	13,147	7.5	63,459	9.8	20,878	8.8	12,474	7.6	60,064	9.9	19,508	8.9
2011	17,043	9.7	64,517	9.9	21,872	9.2	16,030	9.8	60,481	10.0	20,215	9.3
2012	16,883	9.6	55,385	8.5	20,372	8.6	15,710	9.6	51,506	8.5	18,601	8.5
2013	14,638	8.3	44,842	6.9	17,920	7.5	13,278	8.1	40,698	6.7	16,091	7.4
2014	14,951	8.5	41,848	6.5	17,107	7.2	13,513	8.3	37,631	6.2	15,175	6.9
2015	17,701	10.1	42,058	6.5	18,570	7.8	15,543	9.5	36,450	6.0	15,880	7.3
2016	13,315	7.6	26,949	4.2	12,254	5.2	12,317	7.6	24,834	4.1	11,221	5.1
2017	7591	4.3	15,116	2.3	8269	3.5	7039	4.3	14,000	2.3	7615	3.5
2018	3981	2.3	7566	1.2	4702	2.0	3751	2.3	7116	1.2	4367	2.0
2019	3362	1.9	5965	0.9	3680	1.6	3143	1.9	5587	0.9	3411	1.6
2020	1117	0.6	2007	0.3	1155	0.5	1058	0.7	1869	0.3	1063	0.5
Pain type
Chronic back pain	83,896	47.8	261,191	40.3	86,418	36.4	77,782	47.7	242,730	40.1	78,610	36.0
Chronic neck pain	31,316	17.8	104,553	16.1	36,273	15.3	29,094	17.8	96,964	16.0	32,898	15.1
Chronic osteoarthritis	123,002	70.1	483,379	74.5	187,565	78.9	114,582	70.2	451,760	74.7	173,105	79.2
Neuropathic pain	11,135	6.3	30,460	4.7	12,697	5.3	10,297	6.3	27,967	4.6	11,448	5.2
Migraine	9021	5.1	24,634	3.8	9973	4.2	8174	5.0	22,314	3.7	8977	4.1
Abdominal pain	24,507	14.0	88,467	13.6	40,327	17.0	21,916	13.4	79,472	13.1	35,888	16.4
Chest pain	18,393	10.5	61,188	9.4	26,020	11.0	15,915	9.8	53,109	8.8	22,345	10.2
Other pain conditions	33,555	19.1	142,546	22.0	59,955	25.2	30,734	18.8	131,749	21.8	54,976	25.2
Mental health disorders
Major depressive disorder	7748	4.4	23,318	3.6	9640	4.1	6482	4.0	19,744	3.3	7992	3.7
Anxiety disorders	16,968	9.7	53,675	8.3	21,511	9.1	14,193	8.7	45,494	7.5	17,782	8.1
Mood disorders	3568	2.0	10,188	1.6	3817	1.6	2557	1.6	7552	1.3	2715	1.2
Personality disorders	677	0.4	1707	0.3	835	0.4	491	0.3	1220	0.2	604	0.3
Posttraumatic stress disorder	1204	0.7	3229	0.5	1472	0.6	867	0.5	2414	0.4	1099	0.5
Nicotine dependence	12,363	7.0	45,722	7.1	21,375	9.0	2514	1.5	9173	1.5	6136	2.8
Schizophrenia	620	0.4	1309	0.2	444	0.2	360	0.2	793	0.1	238	0.1
Any surgery	103,869	59.2	422,835	65.2	183,373	77.1	96,172	58.9	392,826	64.9	168,180	76.9
Obesity	17,248	9.8	46,560	7.2	22,183	9.3	15,651	9.6	42,081	7.0	19,939	9.1
Charlson Comorbidity Index
Mean (SD)	0.5 (1.0)		0.4 (0.9)		0.5 (1.0)		0.5 (1.0)		0.4 (0.9)		0.4 (1.0)
0	125,474	71.5	498,212	76.8	175,200	73.6	118,213	72.4	469,729	77.6	163,408	74.7
1	33,187	18.9	106,589	16.4	42,111	17.7	30,132	18.5	96,629	16.0	37,711	17.3
2	7883	4.5	21,737	3.4	9641	4.1	6980	4.3	18,999	3.1	8278	3.8
3 or more	8959	5.1	22,364	3.5	10,810	4.6	7869	4.8	19,634	3.3	9227	4.2
Baseline drug use
Benzodiazepines	20,473	11.7	77,542	12.0	31,545	13.3	18,346	11.2	70,009	11.6	28,058	12.8
Nonbenzodiazepine hypnotics	9235	5.3	32,216	5.0	12,259	5.2	8615	5.3	30,102	5.0	11,298	5.2
Gabapentinoids	12,040	6.9	23,403	3.6	10,581	4.5	10,717	6.6	20,552	3.4	9065	4.2
Skeletal muscle relaxants	49,926	28.5	137,889	21.3	39,097	16.4	46,105	28.3	127,569	21.1	35,459	16.2
Antidepressants	38,324	21.8	129,025	19.9	45,159	19.0	34,391	21.1	116,828	19.3	40,008	18.3

Table 3 describes the events observed, the person time, and the unadjusted event rates expressed in events per 100,000 person-years for each AE for each of the 13 exposure groups observed over the follow-up period. A total of 2.9 million person-years of follow-up were available for the opioid-adverse cohort and 2.7 million person-years of follow-up were available for the substance use disorder cohort. A total of 1443 and 4806 individuals had opioid overdose based on the main and the broadest definitions respectively. Accident-related injuries were the most common AE, with 103,172 individuals having the outcome and opioid use disorder was detected in 9135 persons. The average daily MME of the low-dose tramadol, short-acting hydrocodone, and short-acting oxycodone groups were ~19, 25, and 31 MME and were 63, 64, and 77 MME for the high-dose groups, respectively (Table 4; Supplement Table 3–9, Supplemental Digital Content 3, https://links.lww.com/CJP/A903, Supplemental Digital Content 4, https://links.lww.com/CJP/A904, Supplemental Digital Content 5, https://links.lww.com/CJP/A905, Supplemental Digital Content 6, https://links.lww.com/CJP/A906, Supplemental Digital Content 7, https://links.lww.com/CJP/A907, Supplemental Digital Content 8, https://links.lww.com/CJP/A908, Supplemental Digital Content 9, https://links.lww.com/CJP/A909).

TABLE 3 - Outcome Rates for the Time-varying Opioid Exposure Groups

Exposure group	Events; person-years (percent of total follow-up time)	Event/100,000 person-years	Events; person-years (percent of total follow-up time)	Event/100,000 person-years
	Opioid overdose (main measure)		Opioid overdose (broad measure 1)*
No opioids	829; 2,698,741 (93.6)	30.7	1177; 2,697,661 (93.6)	43.6
Low-dose short-acting hydrocodone only	160; 78,550 (2.7)	203.7	210; 78,509 (2.7)	267.5
Low-dose tramadol only	69; 44,390 (1.5)	155.4	108; 44,364 (1.5)	243.4
Low-dose short-acting oxycodone only	83; 20,485 (0.7)	405.2	103; 20,467 (0.7)	503.3
Low-dose tramadol combination	11; 2766 (0.1)	397.8	18; 2764 (0.1)	651.3
Low-dose nontramadol combination	<5; 1001 (0.0)	Not calculated	5; 1000 (0.0)	500.1
Low dose other opioids	30; 5889 (0.2)	509.4	43; 5879 (0.2)	731.5
High-dose short-acting hydrocodone only	20; 6052 (0.2)	330.5	26; 6049 (0.2)	429.8
High-dose tramadol only	<5; 292 (0.0)	Not calculated	<5; 292 (0.0)	Not calculated
High-dose short-acting oxycodone only	62; 11,131 (0.4)	557.0	90; 11,115 (0.4)	809.7
High-dose tramadol combination	18; 2001 (0.1)	899.8	26; 1999 (0.1)	1300.5
High-dose nontramadol combination	95; 6857 (0.2)	1385.4	114; 6844 (0.2)	1665.6
High dose other opioids	62; 6680 (0.2)	928.2	75; 6668 (0.2)	1124.7
Total	1443; 2,884,835	50.0	1999; 2,883,611	69.3
	Opioid overdose (broad measure 2)†		Accidents
No opioids	3416; 2,693,019 (93.6)	126.9	88,265; 2,361,665 (93.5)	3737.4
Low-dose short-acting hydrocodone only	375; 78,306 (2.7)	478.9	6207; 69,476 (2.8)	8934.0
Low-dose tramadol only	198; 44,263 (1.5)	447.3	2413; 39,812 (1.6)	6060.9
Low-dose short-acting oxycodone only	186; 20,395 (0.7)	912.0	1708; 17,733 (0.7)	9631.9
Low-dose tramadol combination	31; 2752 (0.1)	1126.6	322; 2347 (0.1)	13,718.7
Low-dose nontramadol combination	14; 996 (0.0)	1406.3	180; 848 (0.0)	21,218.9
Low dose other opioids	59; 5848 (0.2)	1008.9	451; 4894 (0.2)	9216.2
High-dose short-acting hydrocodone only	49; 6037 (0.2)	811.7	808; 5281 (0.2)	15,299.3
High-dose tramadol only	<5; 291 (0.0)	Not calculated	33; 268 (0.0)	12,327.3
High-dose short-acting oxycodone only	154; 11,059 (0.4)	1392.6	1244; 9560 (0.4)	13,012.2
High-dose tramadol combination	37; 1988 (0.1)	1861.4	238; 1695 (0.1)	14,038.3
High-dose nontramadol combination	166; 6781 (0.2)	2448.1	768; 5590 (0.2)	13,737.9
High dose other opioids	117; 6616 (0.2)	1768.5	535; 5,629 (0.2)	9504.5
Total	4806; 2,878,351	167.0	103,172; 2,524,798	4086.4
	Self-inflicted injuries and poisoning		Violence-related injuries
No opioids	2237; 2,695,997 (93.6)	83.0	1837; 2,695,208 (93.5)	68.2
Low-dose short-acting hydrocodone only	237; 78,443 (2.7)	302.1	138; 78,502 (2.7)	175.8
Low-dose tramadol only	95; 44,333 (1.5)	214.3	32; 44,392 (1.5)	72.1
Low-dose short-acting oxycodone only	100; 20,441 (0.7%)	489.2	44; 20,473 (0.7%)	214.9
Low-dose tramadol combination	12; 2758 (0.1)	435.1	12; 2759 (0.1)	435.0
Low-dose nontramadol combination	6; 1001 (0.0)	599.1	5; 1003 (0.0)	498.6
Low dose other opioids	31; 5866 (0.2)	528.5	7; 5902 (0.2)	118.6
High-dose short-acting hydrocodone only	21; 6039 (0.2)	347.8	12; 6056 (0.2)	198.2
High-dose tramadol only	0; 292 (0.0)	0	0; 292 (0.1)	0
High-dose short-acting oxycodone only	65; 11,104 (0.4)	585.4	28; 11,153 (0.4)	251.1
High-dose tramadol combination	21; 1998 (0.1)	1051.1	<5; 2010 (0.1)	Not calculated
High-dose nontramadol combination	81; 6834 (0.2)	1185.2	9; 6930 (0.2)	129.9
High dose other opioids	52; 6659 (0.2)	780.9	9; 6746 (0.2)	133.4
Total	2958; 2,881,765	102.7	2136; 2,881,426	74.1
	Opioid use disorder		Nonopioid substance use disorder
No opioids	5618; 2,540,315 (93.9)	221.2	73023; 2,363,740 (94.2)	3089.3
Low-dose short-acting hydrocodone only	1159; 69,648 (2.6)	1664.1	6864; 61,361 (2.5)	11,186.2
Low-dose tramadol only	336; 40,710 (1.5)	825.3	2503; 37,413 (1.5)	6690.2
Low-dose short-acting oxycodone only	486; 17,657 (0.7)	2752.4	1,919; 15,380 (0.6)	12,477.3
Low-dose tramadol combination	60; 2398 (0.1)	2,501.7	302; 2060 (0.1)	14,658.7
Low-dose nontramadol combination	35; 835 (0.0)	4,191.9	163; 688 (0.0)	23,698.3
Low dose other opioids	136; 5100 (0.2)	2666.9	411; 4429 (0.2)	9280.1
High-dose short-acting hydrocodone only	129; 5359 (0.2)	2407.1	731; 4727 (0.2)	15,463.6
High-dose tramadol only	6; 264 (0.0)	2271.0	33; 243 (0.0)	13,559.6
High-dose short-acting oxycodone only	442; 9149 (0.3)	4830.9	1365; 7861 (0.3)	17,363.8
High-dose tramadol combination	85; 1712 (0.1)	4965.1	291; 1407 (0.1)	20,681.1
High-dose nontramadol combination	406; 5367 (0.2)	7564.2	845; 4309 (0.2)	19,608.6
High dose other opioids	237; 5577 (0.2)	4249.5	593; 4813 (0.2)	12,320.5
Total	9135; 2,704,091	337.8	89,043; 2,508,740	3549.8

^*Opioid overdose diagnosis codes (main overdose definition) + Naloxone administration in the emergency department + Opioid-related adverse effects codes.

^†Opioid overdose diagnosis codes (main overdose definition) + Naloxone administration in the emergency department + Opioid-related adverse effects codes+Respiratory depression (inpatient/outpatient/emergency department; age 50 years or less) + Central nervous system depression (inpatient/outpatient/emergency department; age 50 years or less).

TABLE 4 - Summary Statistics for Average Daily MME for the Different Dose-based Exposure Groups for the Main Overdose Measure

Exposure groups	Mean	Median	SD	Minimum	Maximum
Low-dose short-acting hydrocodone only	24.9	25.0	10.0	0	50.0
Low-dose tramadol only	18.7	16.4	8.7	0	50.0
Low-dose short-acting oxycodone only	30.7	32.8	11.0	0	50.0
Low-dose tramadol combination	29.3	30.4	12.6	0	50.0
Low-dose nontramadol combination	31.9	33.8	12.8	0	50.0
Low dose other opioids	24.7	22.5	12.2	0	50.0
High-dose short-acting hydrocodone only	64.3	56.3	30.9	50.0	7500.0
High-dose tramadol only	63.6	50.0	35.2	50.0	1000.0
High-dose short-acting oxycodone only	77.3	75.0	41.3	50.0	2857.1
High-dose tramadol combination	90.4	70.0	65.6	50.0	1930.0
High-dose nontramadol combination	140.3	102.3	127.1	50.0	4020.0
High dose other opioids	197.1	171.4	155.7	50.0	6000.0

MME indicates morphine milligrams equivalent.

Compared with no opioid therapy, all the opioid therapy groups were associated with increased risks of all study outcomes in the adjusted and unadjusted analyses except for the outcome of violence-related injuries (Fig. 1; Supplement Tables 10, Supplemental Digital Content 10, https://links.lww.com/CJP/A910 and 11, Supplemental Digital Content 11, https://links.lww.com/CJP/A911). For violence-related injuries, no difference in risk was observed between no opioid therapy and the exposure groups of low-dose tramadol only, high-dose tramadol combination, high-dose nontramadol combination, low-dose other opioids, and high-dose other opioids. In general, the high-dose opioid exposure groups had higher risks of opioid overdose and opioid use disorder than their lower-dose exposure counterparts. For example, high-dose hydrocodone therapy (compared with no opioid therapy) had a higher risk of opioid use disorder (HR [95% CI]: 10.48 [8.79-12.49]) than low-dose hydrocodone therapy (compared with no opioid therapy) (HR [95% CI]: 6.05 [5.66-6.47]) (Fig. 1; Supplement Table 11, Supplemental Digital Content 11, https://links.lww.com/CJP/A911).

FIGURE 1: Adjusted hazard ratios (HRs) (95% CI) for dose-based opioid exposure categories (reference = no opioids). The models were adjusted for age, sex, region of residence, insurance type, index year, chronic back pain, chronic neck pain, chronic osteoarthritis, neuropathic pain, migraine, abdominal pain, chest pain, other miscellaneous pain, major depressive disorder, anxiety disorders, mood disorders, personality disorders, post-traumatic stress disorder, nicotine dependence, Schizophrenia, any surgical procedure, obesity, Charlson comorbidity index, state drug overdose rate (index year), benzodiazepines, non-benzodiazepine hypnotics, gabapentinoids, skeletal muscle relaxants and antidepressants; Opioid overdose (broad measure 1): Opioid overdose diagnosis codes (main overdose definition) + Naloxone administration in the Emergency Department + Opioid-related adverse effects codes; Opioid overdose (broad measure 2): Opioid overdose diagnosis codes (main overdose definition) + Naloxone administration in the Emergency Department + Opioid-related adverse effects codes + Respiratory depression (inpatient/outpatient/Emergency Department; age 50 years or less) + Central Nervous System depression (inpatient/outpatient/Emergency Department; age 50 years or less). There were zero events in the high dose tramadol group for self-inflicted injuries and violence-related injuries, so the regression coefficients were not estimated.

In the adjusted analyses comparing the exposure groups to low-dose hydrocodone-only therapy, low-dose short-acting oxycodone only had a higher risk of opioid overdose (main overdose measure: HR [95% CI]: 1.79 [1.37-2.33]), self-inflicted injuries (HR [95% CI]: 1.47 [1.16-1.86]), opioid use disorder (HR [95% CI]: 1.47 [1.33-1.64]), and nonopioid substance use disorder (HR [95% CI]: 1.13 [1.07-1.19]) (Fig. 2; Supplement Table 11, Supplemental Digital Content 11, https://links.lww.com/CJP/A911). Low-dose tramadol only was not associated with opioid overdose (main overdose measure: HR [95% CI]: 0.85 [0.64-1.13]) but was associated with a lower risk of accidents (HR [95% CI]: 0.73 [0.70-0.77]), self-inflicted injuries (HR [95% CI]: 0.75 [0.59-0.95]), and violence-related injuries (HR [95% CI]: 0.55 [0.37-0.80]), compared with low-dose hydrocodone only. In general, the risk of all the outcomes was higher among the high-dose groups compared with the low-dose groups with the highest risks observed for the high-dose groups, in which opioids were combined. For example, high-dose opioid combinations not involving tramadol had a substantially elevated risk of opioid overdose (main overdose measure: HR [95% CI]: 6.76 [5.22-8.76]), self-inflicted injuries [HR, 95% CI: 3.63 [2.82–4.68]), and opioid use disorders (HR [95% CI]: 4.38 [3.91–4.91]) compared with low-dose hydrocodone only.

FIGURE 2: Adjusted HRs (95% confidence intervals) for dose-based exposure categories (reference = low-dose short-acting hydrocodone only).

Sensitivity Analyses

In the sensitivity analysis using the actual days in each exposure group, low-dose oxycodone-only therapy had higher risks of opioid overdose (main overdose measure: adjusted HR [95% CI]: 1.84 [1.36-2.48]), self-inflicted injuries (adjusted HR [95% CI]: 1.46 [1.12-1.89]), opioid use disorder (adjusted HR [95% CI]: 1.56 [1.38-1.76]), and nonopioid substance use disorder (adjusted HR [95% CI]: 1.14 [1.08-1.21]), compared with low-dose hydrocodone only (Supplement Tables 12, Supplemental Digital Content 12, https://links.lww.com/CJP/A912 and 13, Supplemental Digital Content 13, https://links.lww.com/CJP/A913). No difference in risk for the 3 measures of overdose was observed between low-dose tramadol-only and low-dose hydrocodone-only groups, whereas the low-dose tramadol group had a lower risk of the other outcomes. Similar to the main analysis, high-dose and combination groups had elevated risks of nearly all outcomes, which was more pronounced for the overdose and opioid use disorder outcomes.

In the intent-to-treat analyses, short-acting oxycodone-only initiators had a higher risk of opioid overdose and other adverse effects, (main overdose measure: HR [95% CI]: 1.62 [1.37-1.91]), accidents (HR [95% CI]: 1.08 [1.06-1.10]), self-inflicted injuries (HR [95% CI]: 1.42 [1.25-1.61]), and opioid use disorder (HR [95% CI]: 1.37 [1.28-1.47]) compared with short-acting hydrocodone-only initiators (Supplement Tables 14, Supplemental Digital Content 14, https://links.lww.com/CJP/A915 and 15, Supplemental Digital Content 15, https://links.lww.com/CJP/A916). Initiating opioid therapy on tramadol was not associated with opioid overdose (HR [95% CI]: 1.08 [0.89-1.31]) but was associated with a lower risk of accidents (HR [95% CI]: 0.77 [0.75-0.79]), violence-related injuries (HR [95% CI]: 0.75 [0.64-0.88]) and opioid use disorder (HR [95% CI]: 0.82 [0.76-0.89]) compared with short-acting hydrocodone initiators. Qualitatively similar results were observed with propensity score-adjusted analyses. A modest 4% increase in the hazard of the falsification outcome of atopic dermatitis was observed for tramadol-only initiators (HR [95% CI]: 1.04 [1.01-1.08]), whereas short-acting oxycodone-only initiators had a 6% lower hazard (HR [95% CI]: 0.94 [0.92-0.97]), both compared with short-acting hydrocodone-only initiators (Supplement Table 15, Supplemental Digital Content 15, https://links.lww.com/CJP/A916).

DISCUSSION

Compared with exposure windows where opioids were not dispensed, any opioid therapy at any dose was shown to substantially increase the risk of opioid overdose with HRs of ~3.4 to 9.4 when opioid doses were <50 MME whereas HRs varied between 6.6 and 26.7 for opioid doses ≥50 MME. Similar but less pronounced relationships between dose groups were observed for accidents, self-inflicted injuries, violence-related injuries, opioid use, and nonopioid use disorders. The monotonic increase in opioid overdose and opioid-related mortality with increases in dose is well-documented.^30,38–40 Opioid doses are often titrated based on patients’ adequacy of pain relief; however, escalation of opioid dose is correlated with elevated risks of overdose, injuries, and opioid use disorder, without any improvement in pain intensity.^23,41 Our findings show that the dose-effect is observed across different opioid drugs, emphasizing that prescribers should be cognizant of the risk of high-dose opioid prescriptions irrespective of the type of prescribed opioid.

Regarding one of the main questions this study sought to address, we found that low-dose short-acting oxycodone therapy nearly doubles the risk of opioid overdose (HR: 1.79; 95% CI: 1.37-2.33) compared with low-dose short-acting hydrocodone, whereas the risk of overdose was not significantly different between low-dose tramadol and low-dose short-acting hydrocodone. Low-dose oxycodone was also associated with a higher risk of self-inflicted injuries, whereas low-dose tramadol was associated with modest reductions in the risk of accidents, self-inflicted, and violence-related injuries compared with low-dose hydrocodone. These data seem to suggest that the type of opioid may influence overdose risk, opioid-related accidents, and injuries independent of the opioid dose, such that oxycodone may increase risk whereas tramadol may have an equal or lower risk compared with hydrocodone. Similarly, a study conducted in Tennessee Medicaid adolescents reported a 68% higher risk of opioid overdose for short-acting oxycodone therapy compared with short-acting hydrocodone.⁴² That study, unlike our findings, showed that tramadol had a 185% higher risk of opioid overdose.⁴² The study was conducted at the prescription level, wherein the total days’ supply for each opioid drug was counted without accounting for combination opioid treatment, and the study cohort included both incident and prevalent users, which could partly explain the discrepancy with our findings. Another study conducted using Veterans Affairs data that used a time-varying approach to construct exposure groups reported that tramadol and short-acting opioid therapy conferred similar opioid overdose risk compared with no opioid (tramadol only HR: 1.52; other short-acting opioid HR: 1.69).⁴³ Another study, conducted among Medicare osteoarthritis patients, reported slightly higher adverse outcomes-related hospitalization with nontramadol opioids compared with tramadol therapy. Despite constructing low and high-dose categories in our study, modest differences in dose might also explain the differences in risk between the low-dose opioid therapies as the average daily MME for low-dose short-acting oxycodone only was slightly higher (31 MME) than low-dose short-acting hydrocodone only (25 MME), which was higher than low-dose tramadol only (19 MME).

Similar findings were observed when contrasting the risk of opioid and nonopioid substance use disorders among exclusive therapy of low-dose hydrocodone, oxycodone, and tramadol. Our data suggest that low-dose oxycodone was associated with an increase in the risk of opioid use disorder by ~50% and a smaller increase in nonopioid use disorders whereas low-dose tramadol was associated with a lower risk of these substance use disorders, including opioid use disorders, compared with exclusive use of hydrocodone. A study published using National Survey on Drug Use and Health reported that past-year misuse was 4% among tramadol users, compared with 8% for hydrocodone and oxycodone users, which is consistent with our tramadol findings but at odds with the oxycodone findings having the highest risk of substance use disorders based on documented diagnoses.⁴⁴ Another study comparing self-reported abuse reported similar rates of opioid abuse among tramadol and oxycodone users.⁴⁵ These studies used cross-sectional approaches and self-reported measures of misuse and abuse, which reported descriptive findings compared with our study, which utilized provider-diagnosed cases of substance abuse.^44,45 In addition, these studies compare tramadol abuse and oxycodone abuse, which is a different question than the risk of opioid abuse when prescribed tramadol or oxycodone. Our study could not evaluate the use of illicit opioids among the individuals on tramadol, hydrocodone, or oxycodone therapy. The fact that the risk of nonopioid substance use disorder was also highest among those on low-dose oxycodone therapy (compared with low-dose hydrocodone and tramadol) suggests that illicit drug use might be more prevalent during oxycodone therapy.

The similar or lower rates of opioid-related adverse outcomes and substance use disorders among individuals on low-dose tramadol-only therapy, compared with those on low-dose short-acting hydrocodone only, should be interpreted in the broader context of tramadols’ unique safety profile. Tramadol therapy has been found to be associated with higher risks of hypoglycemia-related hospitalizations, serotonin syndrome and seizures, cardiovascular events, and all-cause mortality, which were risks this study did not consider.^46–48 In addition, individuals initiating opioid therapy on tramadol had a 41% higher risk of long-term opioid use compared with those initiating the therapy on other short-acting opioids.⁴⁹ Tramadol prescribing has increased monotonically in recent years with corresponding decreases in hydrocodone and oxycodone prescribing.^50,51 Although our findings suggest that tramadol has similar risks of overdose and lower risk of accidents, injuries, and substance use disorders compared with hydrocodone, one needs to consider these other risks, and in particular potential increases in all-cause mortality when considering tramadol as an analgesic strategy.

The highest risks of opioid overdoses and opioid use disorders were observed when persons combined opioids leading to high doses or used high-dose opioids that did not include one of the 3 target opioids. In a study conducted using Medicare data on individuals with long-term opioid therapy, combination opioid therapy had a 5-fold increase in the odds of opioid overdose compared with short-acting opioids.⁵² In comparison to no opioid therapy, a combination of short-acting and long-acting was associated with a 384% increase in the hazard of overdose, whereas short-acting only had a 69% higher hazard of overdose.⁴³ Likewise, the literature is consistent on elevated risks of opioid-related adverse outcomes when prescribing long-acting opioids, which accounted for 40% of the high-dose other opioid categories in our study.^43,52,53 Our study reported higher risks of opioid-related adverse outcomes for the combination of opioids with and without tramadol, although the estimates had higher magnitudes for nontramadol combination. These findings suggest that more caution is needed when coprescribing more than 1 opioid, and strategies such as opioid rotation, which is shown to help with pain control in cancer patients,⁵⁴ may be evaluated for chronic noncancer pain.⁵⁵

This study has several limitations. First, although we constructed exposure groups using a time-varying approach, which restricts the attribution of outcome events to the treatment windows, we only adjusted for potential confounders from the baseline period and did not adjust for time-varying confounding. Second, patient characteristics such as family income, race, and pain severity, which are correlates of opioid-related adverse outcomes and could also be associated with the type of opioid prescribed, and prescriber characteristics (specialty, preferences) were not adjusted for in the analyses. Furthermore, confounding by indication is likely as different opioids may be prescribed with different treatment goals. Third, the study could not account for AEs that did not lead to a health care encounter. This could have biased the results if, for instance, individuals on low-dose tramadol were more likely to have AEs that did not involve emergency room visits, outpatient visits, or hospitalizations, compared with those on low-dose hydrocodone. Fourth, the severity of the outcomes was not considered, including whether the events were fatal or nonfatal, due to the lack of death information in the data. This could be meaningful if there is a differential severity of the outcomes across the exposure groups. Lastly, exclusive use of high-dose tramadol is rare. High-dose tramadol only accounted for 0.01% of observed person time, which resulted in less precise estimates for that exposure group and for self-inflicted and violence-related injuries, HRs could not be estimated at all.

In summary, this study found that exclusive use of low-dose oxycodone was associated with increased risks of opioid overdose, violence and self-inflicted injuries, and opioid and nonopioid substance use disorders controlling for MME dose compared with low-dose short-acting hydrocodone among noncancer patients with chronic back or joint pain. Exclusive use of hydrocodone or oxycodone at doses >50 MME was associated with even higher risks of these opioid-related adverse outcomes. Low-dose tramadol was associated with lower risks of opioid and nonopioid substance use disorders compared with low-dose hydrocodone. The highest risk of nearly all the outcomes examined was found among persons combining opioids that had doses >50 MME. These findings confirm the well-established risks of prescribing higher opioid doses and suggest that tramadol at low doses has a lower risk of opioid-related adverse outcomes than low-dose oxycodone. This study did not evaluate the analgesic benefits among the 3 most commonly prescribed opioids and further research is warranted to fully elucidate the risk-benefit profiles of these opioid analgesics. Likewise, the lower risks of low-dose tramadol therapy should be considered together with its serotonin-related AEs.