Association of Tramadol With All-Cause Mortality Among Patients With Osteoarthritis

Key Points

Question  Is tramadol prescription associated with a higher risk of all-cause mortality than other pain relief medications among patients with osteoarthritis?

Findings  In this cohort study that included 88 902 patients with osteoarthritis, initial prescription of tramadol was associated with a significantly increased risk of mortality over 1 year compared with initial prescription of naproxen (hazard ratio [HR], 1.71), diclofenac (HR, 1.88), celecoxib (HR, 1.70), and etoricoxib (HR, 2.04), but not compared with codeine (HR, 0.94).

Meaning  Tramadol prescription may be associated with increased all-cause mortality compared with commonly prescribed nonsteroidal anti-inflammatory drugs, but further research is needed to determine if this relationship is causal.

Importance  An American Academy of Orthopaedic Surgeons guideline recommends tramadol for patients with knee osteoarthritis, and an American College of Rheumatology guideline conditionally recommends tramadol as first-line therapy for patients with knee osteoarthritis, along with nonsteroidal anti-inflammatory drugs.

Objective  To examine the association of tramadol prescription with all-cause mortality among patients with osteoarthritis.

Design, Setting, and Participants  Sequential, propensity score–matched cohort study at a general practice in the United Kingdom. Individuals aged at least 50 years with a diagnosis of osteoarthritis in the Health Improvement Network database from January 2000 to December 2015, with follow-up to December 2016.

Exposures  Initial prescription of tramadol (n = 44 451), naproxen (n = 12 397), diclofenac (n = 6512), celecoxib (n = 5674), etoricoxib (n = 2946), or codeine (n = 16 922).

Main Outcomes and Measures  All-cause mortality within 1 year after initial tramadol prescription, compared with 5 other pain relief medications.

Results  After propensity score matching, 88 902 patients were included (mean [SD] age, 70.1 [9.5] years; 61.2% were women). During the 1-year follow-up, 278 deaths (23.5/1000 person-years) occurred in the tramadol cohort and 164 (13.8/1000 person-years) occurred in the naproxen cohort (rate difference, 9.7 deaths/1000 person-years [95% CI, 6.3-13.2]; hazard ratio [HR], 1.71 [95% CI, 1.41-2.07]), and mortality was higher for tramadol compared with diclofenac (36.2/1000 vs 19.2/1000 person-years; HR, 1.88 [95% CI, 1.51-2.35]). Tramadol was also associated with a higher all-cause mortality rate compared with celecoxib (31.2/1000 vs 18.4/1000 person-years; HR, 1.70 [95% CI, 1.33-2.17]) and etoricoxib (25.7/1000 vs 12.8/1000 person-years; HR, 2.04 [95% CI, 1.37-3.03]). No statistically significant difference in all-cause mortality was observed between tramadol and codeine (32.2/1000 vs 34.6/1000 person-years; HR, 0.94 [95% CI, 0.83-1.05]).

Conclusions and Relevance  Among patients aged 50 years and older with osteoarthritis, initial prescription of tramadol was associated with a significantly higher rate of mortality over 1 year of follow-up compared with commonly prescribed nonsteroidal anti-inflammatory drugs, but not compared with codeine. However, these findings may be susceptible to confounding by indication, and further research is needed to determine if this association is causal.

Few safe and effective treatments are available for patients with osteoarthritis. The main goal of medical therapy for managing osteoarthritis is to control pain while avoiding therapeutic toxicity.¹ Tramadol, a weak opioid agonist, has been considered a potential alternative to traditional opioid agonists in managing pain.² Current American Academy of Orthopaedic Surgeons guidelines strongly recommended tramadol or nonsteroidal anti-inflammatory drugs (NSAIDs) for symptomatic knee osteoarthritis.³ The most recent American College of Rheumatology guidelines (from 2012) conditionally recommended tramadol as a first-line therapy for patients with knee osteoarthritis, along with NSAIDs.⁴ Tramadol prescription for management of knee osteoarthritis doubled from 5% to 10% from 2003 to 2009 in the United States.⁵

A meta-analysis showed no statistically significant association of tramadol vs NSAIDs for pain relief among patients with osteoarthritis,⁶ but tramadol was associated with more opioid-related adverse events (eg, nausea, dizziness, constipation, vomiting, somnolence, tiredness, headache).⁷ Few studies have examined the relationship between tramadol prescription and all-cause mortality, and current evidence regarding the association of tramadol with mortality rates compared with other analgesic medications is inconclusive.^8-13 The present study examined the association of initial prescription of tramadol with all-cause mortality compared with alternative commonly prescribed analgesics in patients with osteoarthritis.

The Health Improvement Network (THIN) is an electronic medical record database derived from the records of general practitioners (GPs) in the United Kingdom. THIN contains health information on approximately 11.1 million patients from 580 general practices in the United Kingdom. Health care information is recorded on site at each practice and includes information on sociodemographics, anthropometrics, lifestyle factors, details from GP visits (eg, disease diagnosis, medication prescription), diagnoses from specialists’ referrals and hospital admissions, as well as results of laboratory tests. The Read classification system is used to code specific diagnoses, and a drug dictionary based on data from the Multilex classification system is used to code drugs. The scientific review committee for the THIN database and the institutional review board at Xiangya Hospital approved this study, with waiver of informed consent.

Eligible participants were patients aged 50 years or older with history of knee, hip, or hand osteoarthritis, based on Read codes, who visited the participating GP office between January 2000 and December 2015. All participants had at least 1 year of continuous enrollment with the general practice. Patients with a history of cancer or an opioid use disorder before study entry were excluded.

We conducted 5 sequential propensity score–matched cohort studies to compare all-cause mortality between participants who received an initial prescription of tramadol and participants who received initial prescription of 1 of the following medications: naproxen or diclofenac (commonly prescribed nonselective NSAIDs), celecoxib or etoricoxib (cyclooxygenase 2 [COX-2] inhibitors), or codeine (a commonly prescribed weak opioid). For example, to compare all-cause mortality between tramadol and naproxen, eligible participants were required to be prescribed neither tramadol nor naproxen 1 year before entering the study. The date of initial prescription of either tramadol or naproxen was considered the index date for the corresponding patient. We divided calendar time into 16 1-year blocks from January 2000 to December 2015. Follow-up ended on December 31, 2016. Within each time block, we calculated propensity scores for initial prescription of tramadol using logistic regression. The variables included in the model were sociodemographic factors (ie, age at index date, sex, Townsend Deprivation Index), body mass index (BMI), lifestyle factors (ie, drinking habits and smoking status), osteoarthritis duration, comorbidities and prescriptions prior to the index date, and health care utilization during the 2 years before the index date. Within each time block, tramadol prescriptions were matched 1:1 to naproxen prescriptions using the greedy matching method.¹⁴ We took the same approach to assemble 4 other propensity score–matched cohort studies: tramadol vs diclofenac, tramadol vs celecoxib, tramadol vs etoricoxib, and tramadol vs codeine.

The primary outcome was all-cause mortality 1 year (hereafter referred to as mortality) after initial prescription of tramadol or its comparators, defined by the death date recorded in THIN, linked to the National Health Service. The change in a patient’s vital status to “dead” is immediately updated in the patient’s electronic health record and requires no input by the practice staff in THIN.

We described the annual prevalence and the treatment duration of prescriptions for tramadol, naproxen, diclofenac, celecoxib, etoricoxib, and codeine among patients with osteoarthritis in THIN between 2000 and 2015. We compared the baseline characteristics of the 5 tramadol cohorts with each of the 5 comparison cohorts. For each patient, we calculated person-years of follow-up as the amount of time from the index date to the first of the following events to occur: death, disenrollment from a GP practice participating in THIN (ie, transferring out of the GP practice; approximately 6% of the included individuals), or the end of a 1-year follow-up period. We calculated mortality for each cohort and plotted Kaplan-Meier mortality curves. We compared mortality in the tramadol cohort with each of the 5 comparison cohorts using Cox proportional hazard models adjusted for calendar year. Patients with missing values for BMI, drinking habits, smoking status, or Townsend Deprivation Index were excluded from analysis. We tested the proportional hazards assumption for each comparison cohort using the Kolmogorov supremum test.¹⁵ If the proportional hazard assumption was violated, we estimated the hazard ratio at 3 months, 6 months, 9 months, and 12 months. We also estimated absolute rate differences (RDs) in mortality between the tramadol cohorts and each of the 5 comparative cohorts.

We performed 6 sensitivity analyses to assess the robustness of our study findings. First, we used asymmetric trimming to exclude patients whose propensity score was below the 2.5th percentile of the propensity score of the tramadol cohort and above the 97.5th percentile of the propensity score of the comparator cohort.¹⁶ Second, to minimize residual confounding by indication when comparing mortality between each tramadol cohort with the comparison cohorts, we conducted a stratified analysis according to the prescription of other opioids before initiation of either tramadol or its comparator. Third, to account for nonadherence of medications under investigation during the study period, we conducted an “as-treated” analysis. Specifically, we censored the follow-up at the time when participants either changed (eg, switching from tramadol to naproxen or vice versa, when comparing tramadol with naproxen) or discontinued (ie, no prescription refill for the respective class of medication for more than 60 days) their initiated medication. Fourth, we performed an analysis among participants whose osteoarthritis was diagnosed during the study period (ie, incident osteoarthritis) to minimize potential misclassification of the duration of osteoarthritis. Fifth, because individuals with missing values were not included in our primary analyses, we performed imputation analyses to account for missing data. Specifically, missing values of the variables listed above were imputed by a sequential regression method based on a set of covariates as predictors. To minimize random error, we imputed 5 data sets, calculating effect estimates from each imputed data set and averaging estimates and their CIs obtained from each imputed data set using Rubin’s rules.¹⁷ Sixth, to minimize the potential reverse-causality bias (ie, protopathic bias) we introduced a 6-month or 1-year exposure lag period to account for a potential latency time window (eg, excluding cases of cancer that occurred within 6 months or 1 year).¹⁸

In addition, we compared cause-specific mortality in each tramadol cohort with each matched comparator cohort using a cause-specific Cox-proportional hazard model to account for competing risk of other causes of death. The cause-specific mortality was defined as either data set–documented cause of death or use of a death-attribution algorithm reliant on postmortem or premortem diagnostic codes when there was no documented cause of death.¹⁹

All P values were 2-sided and P < .05 was considered significant for all tests. All statistical analyses were conducted using SAS version 9.4.

After propensity score matching, 88 902 patients were included in the analysis (mean [SD] age, 70.1 [9.5] years; 61.2% were women). Of the matched participants, 12 397 were included in the naproxen cohort, 6512 in the diclofenac cohort, 5674 in the celecoxib cohort, 2946 in the etoricoxib cohort, and 16 922 in the codeine cohort. As shown in Figure 1, the prevalence of participants with knee, hip, or hand osteoarthritis with prescriptions for tramadol increased from 1561 of 46 481 (3.4%) in 2000 to 12 633 of 113 856 (11.1%) in 2013, then decreased to 8407 of 86 014 (9.8%) in 2015. The prevalence of participants with naproxen prescriptions increased from 1830 of 46 481 (3.9%) in 2000 to 11 285 of 86 014 (13.1%) in 2015, whereas diclofenac prescription rates declined from 7512 of 46 481 participants (16.2%) in 2000 to 2161 of 86 014 (2.5%) in 2015. Participants with celecoxib prescriptions increased from 292 of 46 481 (0.6%) in 2000 to 6658 of 75 945 (8.8%) in 2004, then declined after 2005. Etoricoxib entered the UK market in 2002, and the annual prevalence of its prescription remained low during the study period (204 of 62 692 participants [0.3%] in 2002 and 479 of 86 014 [0.6%] in 2015). The prevalence of participants with codeine prescription increased over time from 1497 of 46 481 (3.2%) in 2000 to 4297 of 86 014 (5.0%) in 2015.

The mean (range) treatment duration of a prescription for tramadol was 22 (5-67) days; naproxen, 24 (5-60) days; diclofenac, 24 (5-60) days; celecoxib, 31 (5-60) days; etoricoxib, 27 (5-60) days; and codeine, 25 (5-150) days among patients with osteoarthritis.

As shown in Table 1 and the Supplement, participants in the tramadol cohort, in general, were older; had a higher BMI; had a longer duration of osteoarthritis; and had a higher prevalence of comorbidities (eg, peptic ulcer, chronic kidney disease, diabetes, hypertension, and cardiovascular diseases), other prescriptions (eg, other NSAIDs, other opioids, aspirin, statin, antihypertensive medicine, and antidiabetic medicine), and health care utilization than participants in the NSAIDs cohorts before propensity score matching. After matching, the characteristics between the 2 matched cohorts were well balanced, with all standardized differences less than 0.10 (Table 2).

Mortality was higher in the tramadol cohort than in the naproxen cohort (Figure 2A). During the 1-year follow-up period, 278 deaths (23.5 per 1000 person-years) occurred in the tramadol cohort and 164 deaths (13.8 per 1000 person-years) occurred in the matched naproxen cohort (Table 3). Compared with the naproxen cohort, the RD of mortality for tramadol was 9.7 per 1000 person-years (95% CI, 6.3-13.2). Because the proportional hazard assumption was violated for the comparison of tramadol vs naproxen (P < .001), follow-up time was divided into less than or equal to 3, 6, 9, and 12 months, and the hazard ratios (HR) at 3 months was 2.93 (95% CI, 2.02-4.26), 6 months was 2.34 (95% CI, 1.80-3.05), 9 months was 1.93 (95% CI, 1.55-2.40), and 12 months was 1.71 (95% CI. 1.41-2.07). Tramadol was also associated with higher mortality than diclofenac in the matched cohorts (Figure 2B). Compared with the diclofenac cohort, the RD of mortality for tramadol prescription was 17.0 per 1000 person-years (95% CI, 11.2-22.8) and the HR was 1.88 (95% CI, 1.51-2.35) (Table 3).

Mortality in the tramadol cohort was higher than in the celecoxib cohort (Figure 2C). During the 1-year follow-up period, 171 deaths (31.2 per 1000 person-years) occurred in the tramadol cohort and 102 deaths (18.4 per 1000 person-years) occurred in the celecoxib cohort (Table 3). The RD of mortality for tramadol vs celecoxib was 12.8 per 1000 person-years (95% CI, 6.9-18.7) and the HR was 1.70 (95% CI, 1.33-2.17). Similar findings were observed when mortality in the tramadol cohort was compared with the etoricoxib cohort (Figure 2D and Table 3).

There was no statistically significant difference in mortality between the tramadol cohort and the matched codeine cohort (Figure 2E and Table 3). During the 1-year follow-up period, 519 deaths (32.2 per 1000 person-years) occurred in the tramadol cohort and 552 deaths (34.6 per 1000 person-years) occurred in the codeine cohort. The RD of mortality for tramadol was −2.3 per 1000 person-years (95% CI, −6.3 to 1.7) and the HR was 0.94 (95% CI, 0.83-1.05).

Results from sensitivity analyses also showed that participants in the tramadol cohort experienced significantly higher mortality than those in the naproxen, diclofenac, celecoxib, and etoricoxib cohorts, but not in the codeine cohorts (Table 3).

As shown in Table 4, mortality rates from cardiovascular, gastrointestinal, infection, cancer, and respiratory diseases were higher in the tramadol cohort than in the NSAIDs cohorts; however, because of the relatively small number of deaths from each specific cause, most associations were not statistically significant. No statistically significant difference in each cause-specific morality (except for infection) was observed between the tramadol cohort and the codeine cohort.

Using data collected from THIN, this study found that initial prescription of tramadol was associated with a significantly increased mortality rate over the next year compared with commonly prescribed NSAIDs among participants with osteoarthritis, but no statistically significant difference in mortality rate was observed between tramadol and codeine. Considering that participants with initial prescription of tramadol had a higher comorbidity burden than those with an initial prescription of NSAIDs before propensity score matching, these results were susceptible to confounding by indication. Thus, the present findings should be interpreted with caution, and future studies are needed.

Oral NSAIDs (nonselective NSAIDs and COX-2 inhibitors) are the predominant analgesic medications used to manage osteoarthritis worldwide; however, their safety, particularly with regard to cardiovascular and gastrointestinal risk, has raised concern. Similarly, opioids are commonly prescribed for managing osteoarthritis and their safety has been questioned because of a potential increase in mortality.^20-22 Tramadol is a weak opioid agonist and has been considered a potential alternative to NSAIDs and traditional opioids because of its assumed relatively lower risk of serious cardiovascular and gastrointestinal adverse effects than NSAIDs,²³ as well as a lower risk of addiction and respiratory depression compared with other opioids.² Studies, including the present study, have shown that tramadol prescription among patients with osteoarthritis has been increasing since 2000.⁵

The few studies that have assessed the relationship between tramadol prescription and mortality among patients with different diseases have yielded conflicting results. One study of 1271 patients with perforated peptic ulcer found that tramadol prescription was associated with significantly higher in-hospital mortality than the absence of tramadol or NSAIDs prescription.⁹ Similar results were observed among 153 758 patients receiving dialysis.¹⁰ Furthermore, data on 11.3 million patients from the Clinical Practice Research Datalink showed that the tramadol-related death rate increased before tramadol was classified as a Schedule III controlled substance in 2014 in the United Kingdom, and decreased thereafter.¹³ However, a small study of 272 patients who underwent hip replacement due to fracture showed that tramadol prescription was not associated with increased mortality within 6 months after surgery compared with no prescription of tramadol.¹¹ Another study failed to show a statistically significant mortality difference between prescription of tramadol alone or in combination with codeine compared with infrequent or no prescription of tramadol alone or combined with codeine among 8866 patients with Crohn disease and ulcerative colitis.¹² In a large propensity score–matched cohort study that examined the safety of 5 commonly prescribed opioids among 31 375 Medicare beneficiaries in the United States, initial prescription of tramadol was not associated with a statistically significant higher mortality than hydrocodone prescription after a 180-day follow-up (rate ratio, 1.44 [95% CI, 0.96-2.17]).⁸

The biological mechanisms linking tramadol to mortality are unclear. Tramadol may activate μ opioid receptors and inhibit central serotonin and norepinephrine reuptake, and the latter may result in a unique adverse effect on the neurological system (ie, serotonin syndrome and seizures).² Tramadol may also increase the risk of postoperative delirium, which tends to increase mortality.²⁴ Fatal poisoning or respiratory depression may occur when tramadol users consume alcohol or use tramadol with other central nervous systems depressants.^25-28 Furthermore, tramadol may increase the risk of hypoglycemia, hyponatremia, fracture, or fall, thus leading to an increased risk of death.^29-32

The present findings may have clinical implications. First, if replicated and determined to likely be causal, these findings would indicate an unfavorable safety profile of tramadol. Second, various strategies have been proposed to minimize the adverse effects of analgesics use. For instance, coprescription of proton pump inhibitors with oral NSAIDs has been considered a cost-effective approach for patients with osteoarthritis with moderate or high gastrointestinal risk.^4,33-36 For patients with high cardiovascular risk, naproxen may be preferred, owing to its relatively low cardiovascular risk.^36,37 In general, based on results reported in the current study, nonopioid therapy could be preferred for management of chronic pain (eg, osteoarthritis).³⁸

This study has several limitations. First, 16.4% to 29.7% of causes of death could not be ascertained, and the current study did not have adequate statistical power to evaluate the relationship of initial prescription of tramadol to cause-specific mortality because of a small number of cause-specific deaths. Second, this study found a higher cancer-related mortality in the tramadol cohort than the NSAIDs cohorts. It is possible that some participants were experiencing pain from undetected early-stage cancer and therefore were given stronger pain medication to relieve the symptoms prior to cancer diagnosis (ie, protopathic bias). Although excluding cancer cases that occurred within 6 months or 1 year showed that all-cause mortality in the tramadol cohort was still significantly higher than in 4 NSAIDs cohorts, the increased rate of cancer mortality among patients prescribed tramadol suggests that confounding by indication, such as severity of other comorbidities, may be a potential explanation of the present findings. Third, participants with initial prescription of tramadol were older, had a higher BMI, had a longer duration of osteoarthritis, had a higher prevalence of comorbidities, received more prescriptions, and had more health care utilization than participants in the NSAIDs cohorts before propensity score matching. Thus, while techniques were used to try to control for the potential confounders, including propensity score matching, residual confounding still could affect the study findings. It is possible that comorbidities and illness severity associated with tramadol prescription may explain the higher mortality rate in this group. Fourth, this study was conducted among patients with osteoarthritis. Thus, these findings may not be generalizable to patients with other diseases whose disease pathophysiology may modify the effect of tramadol on mortality.

Among patients aged 50 years and older with osteoarthritis, initial prescription of tramadol was associated with a significantly higher risk of mortality over 1 year of follow-up compared with commonly prescribed NSAIDs, but not when compared with codeine. However, these findings may be susceptible to confounding by indication, and further research is needed to determine if this association is causal.

Corresponding Authors: Guanghua Lei, MD, PhD, Department of Orthopaedics, Xiangya Hospital, Central South University, 87 Xiangya Rd, Changsha, Hunan 410008, China (lei_guanghua@csu.edu.cn); Yuqing Zhang, DSc, Division of Rheumatology, Allergy, and Immunology, Department of Medicine, Massachusetts General Hospital, Harvard Medical School, 55 Fruit St, Boston, MA 02114 (yzhang108@mgh.harvard.edu).

Accepted for Publication: February 5, 2019.

Author Contributions: Dr Zhang had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Drs Lei and Zhang are joint corresponding authors.

Concept and design: Zeng, Dubreuil, Lu, Choi, Lei, Zhang.

Acquisition, analysis, or interpretation of data: Zeng, Dubreuil, Larochelle, Lu, Wei, Lei, Zhang.

Drafting of the manuscript: Zeng, Lu, Choi, Lei, Zhang.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Lu, Wei, Choi, Zhang.

Obtained funding: Zeng, Wei, Choi, Lei, Zhang.

Administrative, technical, or material support: Zeng, Wei, Choi, Lei, Zhang.

Supervision: Choi, Lei, Zhang.

Conflict of Interest Disclosures: Dr Larochelle reported receiving grants from National Institute on Drug Abuse (K23 DA042168) during the conduct of the study and grants from Optum Labs outside the submitted work. No other disclosures were reported.

Funding/Support: This work was supported by the National Institute of Arthritis and Musculoskeletal and Skin Diseases (K23 AR069127, P60 AR047785), and the National Natural Science Foundation of China (81772413, 81702207, 81702206).

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: The interpretation of these data is the sole responsibility of the authors.