OPIOID ANALGESICS AND THE RISK OF SERIOUS INFECTIONS AMONG PATIENTS WITH RHEUMATOID ARTHRITIS: A SELF-CONTROLLED CASE SERIES STUDY

Abstract

Objective

Animal and in vitro studies suggest that certain opioid analgesics impair crucial immune functions. We sought to determine if opioid use is associated with an increased risk of serious infections in patients with rheumatoid arthritis (RA).

Methods

We conducted a self-controlled case series analysis on a retrospective cohort of 13,796 patients with RA enrolled in Tennessee Medicaid (1995–2009). We performed within-person comparisons of the risk of hospitalizations for serious infections during periods of opioid use compared with non-use using conditional Poisson regression. Fixed confounders were accounted for by design, and time-varying confounders included age, the use of disease-modifying anti-rheumatic drugs, glucocorticoids and proton-pump inhibitors. Additional analyses examined new opioid use, use of opioids known to have immunosuppressive properties, long acting opioid use, and different opioid dosages. Sensitivity analyses accounted for potential protopathic bias and confounding by indication.

Results

Among 1,790 patients with RA who had at least one hospitalization for serious infection, the adjusted incidence rate of serious infection was higher during periods of current opioid use compared with non-use [incidence rate ratio (IRR): 1.39 (95% confidence interval (CI): 1.19–1.62)]. The incidence rate was also higher during periods of long-acting opioid use, immunosuppressive opioid use and new opioid use compared with non-use [IRR: 2.01 (95% CI: 1.52–2.66); IRR: 1.72 (95% CI: 1.33–2.23); IRR: 2.38 (95% CI: 1.65–3.42), respectively]. Results of sensitivity analyses were consistent with the main findings.

Conclusions

In within-person comparisons of patients with RA, opioid use was associated with an increased risk of hospitalizations for serious infection.

Patients with rheumatoid arthritis (RA) are at increased risk for serious infections (1–5). This elevated risk is thought to be associated with the autoimmune disease process itself, the impact of comorbidities linked to RA and the use of immunosuppressive medications, including glucocorticoids and disease-modifying anti-rheumatic drugs (DMARDs) (5–11).

Opioid analgesics are increasingly prescribed for patients with chronic non-cancer pain, including patients with RA (12–16). However, the long-term safety of these medications remains unknown (17–23). One concern relates to the immunosuppressive properties of some opioid analgesics. Evidence from both in vitro and animal models indicates that exposure to certain opioids is associated with a reduction in neutrophil and macrophage chemotaxis, as well as the inhibition of complement, natural killer cell and phagocytic activity (24–30). However, the clinical implications of these immunosuppressive properties of opioids remain unclear.

Determining whether the use of opioid analgesics increases the risk of serious infections is of great importance, especially for vulnerable patients with RA. Therefore, the purpose of this study was to compare the risk of serious infections during periods of opioid use compared to periods of non-use among a retrospective cohort of patients with RA enrolled in Tennessee Medicaid (1995–2009). We used a self-controlled case series design (SCCS), in which patients serve as their own controls for within-person comparisons to account directly for the effects of fixed confounders while allowing for control of time-varying factors (31, 32).

MATERIALS AND METHODS

Selection criteria

The Tennessee Medicaid (TennCare) program provides healthcare insurance to those who are Medicaid eligible and to those who are uninsured or otherwise lack access to healthcare. We used TennCare data from 1995 through 2009 supplemented with pharmacy information from Medicare Part D for those that were dual eligible to assemble a retrospective cohort of patients with RA. The study protocol was reviewed and approved with waiver of consent by the Vanderbilt University Institutional Review Board.

Patients became study eligible on the earliest date they fulfilled selection criteria. Patients were identified by a filled DMARD prescription and at least one coded healthcare encounter for RA (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD9-CM]: 714 excluding Juvenile idiopathic arthritis [714.3]) in the 180 days prior to and including the DMARD prescription fill date (33). The DMARDs considered as a qualifying prescription for inclusion in the study population were hydroxychloroquine, leflunomide, methotrexate, sulfasalazine, adalimumab, etanercept, infliximab, cyclophosphamide, cyclosporine, D-penicillamine, gold salts and minocycline. Cohort members were 18 years or older on the date of the first qualifying DMARD prescription, had at least 180 days of continuous enrollment (baseline period), enrolled in a TennCare category with medication benefits or eligible for Medicare Part D (starting in 2006), and had filled at least one prescription during baseline to assure active medical surveillance and use of pharmacy benefits. Additionally, cohort members were free of exposure to non-study opioids (opioids not in Table 1) during baseline and on the first qualifying DMARD prescription fill date and without a hospitalization for any reason during the 30 days preceding the first qualifying DMARD prescription fill date.

Table 1.

Study opioid classifications by name, potency, duration of action, potential for immunosuppression and morphine equivalent conversion factor

Opioid	Potency	Duration of Action	Immuno-suppressive1	Morphine Equivalent Conversion Factor (per mg)2
Hydrocodone3	Moderate	Short-acting	No	1.00
Tramadol4	Moderate	Short-acting	No	0.10
Propoxyphene4	Moderate	Short-acting	Unknown	0.23
Codeine4	Moderate	Short-acting	Yes	0.15
Dihydrocodeine4	Moderate	Short-acting	Yes	0.25
Butalbital/codeine4	Moderate	Short-acting	Yes	0.15
Hydromorphone	Strong	Short-acting	No	4.00
Oxycodone4	Strong	Short-acting	No	1.50
Oxymorphone	Strong	Short-acting	No	3.00
Meperidine4	Strong	Short-acting	Unknown	0.10
Codeine Sulfate5	Strong	Short-acting	Yes	0.15
Morphine Sulfate	Strong	Short-acting	Yes	1.00
Oxycodone6	Strong	Long-acting	No	1.50
Morphine Sulfate7	Strong	Long-acting	Yes	1.00
Methadone Hydrochloride	Strong	Long-acting	Yes	3.00

¹Immunosuppressive categories are based upon the classification for immunosuppression used in a prior study.[34];

²Conversion factors are based on classification used in prior study.[35];

³In combination;

⁴Alone or in combination;

⁵Non-sustained release formulation;

⁶Controlled release formulation;

⁷Sustained release formulation

As certain serious medical conditions could reduce follow-up and increase the risk of serious infection independent of opioid use, we excluded patients with selected conditions during baseline: solid organ transplantation, HIV/AIDS, cancer, serious kidney or liver diseases (Supplement Table 1). As treatment for other non-RA autoimmune diseases could also affect the risk of serious infection, we excluded patients with juvenile idiopathic arthritis, systemic lupus erythematosus, Crohn’s disease, ulcerative colitis, ankylosing spondylitis, or psoriasis/psoriatic arthritis during baseline (Supplement Table 1).

The case-only cohort used in the SCCS analysis included those patients with RA who fulfilled selection criteria and who were hospitalized for serious infection during follow-up and who had at least 2 intervals of follow-up during the study period (to allow for within-person comparisons).

Exposures

The study exposure was the use of any oral opioid analgesic in pill or tablet formulations. We classified all other formulations as non-study opioids. Each person-day of follow-up was classified according to the probability of exposure to the study opioids (Figure 1-A). Current use included any person-time between prescription filling date (inclusive) and the end date of supply. Recent use included up to 180 person-days without opioid exposure, following the last person-day of current use. The recent use category helped to avoid misclassification by allowing for use after the last day of drug supply due to variability in patient adherence to the prescription. Non-use included all person-time that was not current or recent.

Figure 1.

Representation of patient follow-up time in study. A: Representation of follow-up after first qualifying disease-modifying anti-rheumatic (DMARD) prescription fill, and definition of opioid use categories. B: Representation of time during hospitalization and 30-day post-discharge period excluded from follow-up. ¹DMARD: Disease-modifying anti-rheumatic drug.

A subset of current use, defined as new use, was defined as the person-time between the prescription filling date (inclusive) and the end date of supply for the first prescription started after at least 180 days without current opioid use. We further defined individual study opioids according to their potency (moderate or strong), duration of action (short-acting or long-acting), the potential for immunosuppressive effects (known immunosuppression, no immunosuppression or unknown immunosuppression) and according to daily morphine-equivalent dose as defined in the previous literature (Table 1) (34, 35). The quartiles of the daily morphine-equivalent dose distribution were used to create four categories of current use to compare to non-use (0–15mg/day, 15–30mg/day, 30–60mg/day and ≥ 60mg/day).

Follow-up

Follow-up started on the date of the first qualifying DMARD prescription and continued through the earliest of the following: study end date (December 31, 2009), the day prior to the identification of an excluded medical condition (Supplement Table 1), loss of enrollment, the date of death, or the day prior to a prescription fill for a non-study opioid. Since hospitalization-time represents person-time not at risk for the study outcome and medication use during hospitalization is not recorded in our data sources, hospitalization time was excluded from follow-up. Furthermore, the 30-day period following any hospital discharge was excluded, as it is a high-risk period for hospital-acquired infections and directly following a period of unknown medication use (Figure 1-B). However, opioid use post-discharge was considered for classifying opioid use after the end of the 30-day post-discharge period.

Outcome

The outcome of interest was a hospitalization or <24-hour observational stay for a serious infection as identified by a primary discharge diagnosis code for an infection. Specific infections included pneumonia, meningitis, encephalitis, septicemia, cellulitis, soft-tissue infections, endocarditis, pyelonephritis, infective arthritis, and osteomyelitis (Supplement Table 2). Compared with medical chart reviews, the ICD9-CM algorithm used to identify these serious infections was shown to have a positive predictive value greater than 80% for all infection types (36, 37). The date of hospital admission was considered the outcome date. Serious infections occurring within 30 days of the discharge date for a previous infection were considered part of the same episode. Patients could contribute more than one serious infection during the course of follow-up.

Time-varying covariates

Although the effects of all fixed covariates are directly accounted for by the study design (31), we identified a priori time-varying covariates to include in the analysis. Each person-day of follow-up was classified according to the age of the patients and the use of selected medications. Glucocorticoid use was defined as no use, recent use or current use of either a low (≤ 5 mg prednisone equivalents per day), medium (6–14 mg) or high dose (> 14 mg). Other time-varying medications included non-biologic or biologic DMARDs and proton-pump inhibitors (PPI), categorized as no use, recent use or current use. Follow-up was also classified according to nursing home residency status. Since respiratory infections are more prominent during winter, follow-up was further classified as winter (October–March) or summer-related months (April–September).

Statistical analysis

We used a SCCS design to estimate incidence rate ratios (IRR) for serious infections comparing periods of current opioid use to periods of non-use using conditional Poisson regression. Secondary analyses assessed periods of new opioid use compared with non-use, and current use further classified by the duration of opioid action, potential immunosuppressive properties, and daily morphine-equivalent dosage compared with non-use.

Planned sensitivity analyses addressed SCCS assumptions. One assumption is that the outcome does not affect the likelihood of exposure, so we completed sensitivity analyses excluding patients who died during a hospitalization for serious infection, patients who died within 30 days of a hospitalization for serious infection, and any patient who died during follow-up. Another assumption is that events are independent from each other, so two analyses were conducted; one that only included the first serious infection for each patient and another that included only patients with a single infection during follow-up (32).

To address concerns about protopathic bias (e.g. opioid given for a symptom, such as cough, that was an early indicator of pneumonia), we excluded the first 3 person-days of new opioid use in a sensitivity analysis. In addition, we allowed for potential stockpiling of current opioid prescriptions in a separate analysis. As the risk of pneumonia may be related to the respiratory depressive effects of opioids, the risk of pneumonia was assessed separately from other infections in a stratified analysis. To address concerns about the difference in baseline risk among patients receiving strong opioids with those receiving moderate opioids, we compared current opioid use to non-use in patients only receiving opioids of the same potency during follow-up. Finally, to assess for potential confounding by indication (i.e. pain), the association between non-steroidal anti-inflammatory drug (NSAID) use and the risk of serious infection was also assessed. To address the concern for protopathic bias regarding NSAID use, we also excluded the first 3 person-days of new NSAID use in a separate sensitivity analysis. All analyses were completed using STATA, version 14.0 (StataCorp, College Station, TX).

RESULTS

Study Population

We initially identified 13,796 patients with RA who fulfilled selection criteria (Figure 2). As a case-only design, the SCCS cohort included only the 1,790 patients with at least one serious infection (Figure 2). In our study cohort, 2,581 serious infections were identified through 9,686 person-years of follow-up. The reasons for the end of follow-up in this SCCS cohort were the use of a non-study opioid (n=532), death (n=446), end of study (n=316), loss of enrollment (n=284), and identification of an excluded condition (n=212).

Figure 2.

Flow diagram of inclusion/exclusion criteria for self-controlled case series cohort of patients with rheumatoid arthritis and at least one hospitalization for serious infection during follow-up. ¹RA: Rheumatoid arthritis; ²DMARD: Disease-modifying anti-rheumatic drug ³t0: Date of first qualifying disease-modifying anti-rheumatic drug prescription after rheumatoid arthritis diagnosis (start of study follow-up); ⁴SCCS: Self-controlled case series.

The prevalence of at least one period of current opioid use was higher in the SCCS cohort compared to all qualifying patients with RA [95.0% (n=1,701) and 87.3% (n=12,041), respectively]. In the SCCS cohort, 35.8% of follow-up time was classified as current use including hydrocodone (48.0%), propoxyphene (22.3%), oxycodone (12.8%), morphine (5.7%), codeine (5.0%), and other opioids (6.2%). Unadjusted comparisons indicated all selected time-varying covariates, with the exception of DMARD use, were associated with an increase in the risk of infection (Supplement Table 3).

Self-controlled case series analysis

Compared with non-use of opioids and accounting for only fixed confounders, the incidence of serious infections increased during current exposure to opioids [IRR: 1.73 (95% confidence interval (CI): 1.49–2.01] (Table 2). In the fully adjusted analysis accounting for age, season, nursing home residency, and medication use, the incidence of serious infections remained increased during periods of current opioid use compared to non-use [adjusted IRR: 1.39 (95% CI: 1.19–1.62)] (Table 2). Recent opioid use periods had a higher incidence compared to non-use as well [adjusted IRR: 1.40 (95% CI: 1.22, 1.61)]. When examining new opioid use as a subset of current opioid use, the incidence of serious infection was even larger compared with non-use [adjusted IRR: 2.38 (95% CI: 1.65, 3.42)] (Table 2).

Table 2.

Fixed Confounder and Time-Varying Confounder Adjusted Incidence Rate Ratios for Serious Infections by Opioid Exposure Type

Exposure	Fixed Confounder Adjusted IRR1 (95% CI2)	Fixed and Time-Varying Adjusted3 IRR1 (95% CI2)
Opioid Exposure
No Exposure	1.00 (reference)	1.00 (reference)
Current Exposure	1.73 (1.49, 2.01)	1.39 (1.19, 1.62)
New Use4	2.52 (1.75, 3.62)	2.38 (1.65, 3.42)
Duration of Action
No Exposure	1.00 (reference)	1.00 (reference)
Current Short-Acting Opioid	1.66 (1.43, 1.94)	1.35 (1.15, 1.58)
Current Long-Acting Opioid	2.81 (2.14, 3.69)	2.01 (1.52, 2.66)
Immunosuppressive (IS) category
No Exposure	1.00 (reference)	1.00 (reference)
Current Opioid - Unknown IS	1.42 (1.15, 1.75)	1.31 (1.06, 1.62)
Current Opioid - Non-IS	1.81 (1.54, 2.13)	1.37 (1.15, 1.62)
Current Opioid – IS	2.11 (1.64, 2.72)	1.72 (1.33, 2.23)
Daily Morphine Equivalent (MEQ)
No Exposure	1.00 (reference)	1.00 (reference)
Current - <15mg MEQ	1.50 (1.21, 1.86)	1.24 (1.00, 1.54)
Current - 15 – <30mg MEQ	1.76 (1.46, 2.13)	1.40 (1.15, 1.70)
Current - 30 – <60mg MEQ	1.59 (1.31, 1.92)	1.26 (1.04, 1.53)
Current - ≥60mg MEQ	2.15 (1.76, 2.62)	1.73 (1.41, 2.13)

¹IRR: Incidence Rate Ratio;

²CI: Confidence Interval;

³Full model includes age, glucocorticoid use, disease-modifying anti-rheumatic drug use, proton-pump inhibitor use, seasonality and nursing home status;

⁴New use is a subset of current use

Secondary analyses

Secondary analyses explored additional pre-specified stratifications of current opioid use. Periods of current use of long-acting and short-acting opioids both increased the rate of serious infections compared with non-use, with the largest estimate among long-acting opioid use (Table 2). Similarly, use of both immunosuppressive and non-immunosuppressive opioids was associated with an increased risk of infection, although the magnitude was higher among immunosuppressive opioids (Table 2). All three morphine-equivalent dose categories greater than 15mg were associated with an increased risk of infection, with the largest estimate among the highest morphine equivalent category (Table 2).

Sensitivity analyses

Findings from additional analyses designed to address key assumptions of the study design yielded results largely consistent with the main study findings (Table 3). In stratified analyses, current opioid use had a statistically significant association with the incidence rate of non-pneumonia infections, but not with pneumonia (Table 3). Only 2.3% (n=39) of those with current opioid use received only a strong potency opioid during follow-up, precluding further analysis within this group. Among those that did not receive a strong opioid during follow-up (n=887), current use was associated with an increased risk of serious infection compared to non-use (Table 3). However, the precision of these subgroup estimates was limited due to the reduced sample size.

Table 3.

Sensitivity Analyses of the Incidence Rate Ratios of Serious Infections for Current Opioid Use in Comparison to No Opioid Use

Sensitivity Analyses	Patients (Infections)	Current Use Adjusted1 IRR2 (95% CI)
Primary Analysis	1,790 (2,581)	1.39 (1.19, 1.62)
Allowing stockpiling of opioids	1,790 (2,581)	1.45 (1.24, 1.70)
Excluded 3-day period after onset of new use3	1,775 (2,562)	1.34 (1.14, 1.57)
Excluded if death during infection hospitalization	1,774 (2,555)	1.38 (1.18, 1.62)
Excluded if death ≤ 30 days after infection hospitalization	1,639 (2,331)	1.37 (1.16, 1.62)
Excluded if death during follow-up	1,344 (1,883)	1.38 (1.15, 1.66)
Excluded subsequent infections after first infection	1,790 (1,790)	1.44 (1.19, 1.73)
Excluded patients with > 1 infection during follow-up	1,301 (1,301)	1.67 (1.34, 2.09)
Only pneumonia infections during follow-up	1,018 (1,383)	1.22 (0.99, 1.51)
Only non-pneumonia infections during follow-up	597 (709)	1.94 (1.41, 2.67)
No use of strong potency opioids during follow-up	887 (1,218)	1.59 (1.28, 1.98)
Assessing Confounding by Indication
Considering NSAID4 use as primary exposure5	1,790 (2,581)	1.13 (0.99, 1.30)
Excluded 3-day period after onset of new NSAID4 use5	1,783 (2,566)	1.11 (0.96, 1.27)

¹Adjusted for age, season, glucocorticoid use, DMARD use, PPI use, nursing home residency;

²IRR: Incidence rate ratio;

³Fifteen patients were excluded in the sensitivity analysis as the patient’s only infections occurred in the 3-day exclusion period, thereby dropping the patient from the analysis;

⁴NSAID: Non-steroidal anti-inflammatory drug;

⁵Also adjusting for opioid use (no use, recent use, current use)

When considering current NSAID use as the exposure of interest to assess potential confounding by indication (i.e. pain), the rate of hospitalizations for serious infections was not significantly increased compared with periods of non-use [adjusted IRR: 1.13 (95% CI: 0.99, 1.30)] (Table 3).

DISCUSSION

Among patients with RA, the incidence of hospitalizations due to serious infection was higher during periods of new and current opioid use in comparison to periods of non-opioid use. Importantly, the increased risk was observed after adjustment for use of medications known to increase the risk of infections (such as glucocorticoids). Higher risks were associated with long-acting opioids, potentially immunosuppressive opioids, and those with a daily morphine-equivalent dose ≥ 60mg.

The potential association between the risk of infection and opioid use is supported by the literature regarding the immunosuppressive effects of certain opioids from in vitro experiments and animal models, including morphine, methadone and fentanyl (fentanyl not included in our study which is restricted to oral formulations) (19, 26, 38). Specifically concerning morphine, potential mechanisms of immunosuppression include the inhibition of T-cell receptor signaling (24), depletion of lymphocytes (28), and reductions in natural killer cell activity (38, 39). Methadone has been shown to interact with lymphocyte opioid receptors leading to immunosuppression (40), while fentanyl has been linked to both a reduction in natural killer cell activity and reduction in lymphocyte proliferation (25). These same mechanisms have not been demonstrated for other commonly used opioids, such as hydromorphone and oxycodone (19).

Most previous studies between opioid use and the risk of infection involved selected populations of hospitalized or surgical patients, and did not examine community-acquired infections as a primary outcome (41–45). The baseline risk of infection and likelihood for prescription opioid use among patients with RA is potentially different from hospitalized, surgical and osteoarthritis patients that were included in these prior studies. The most directly comparable previous study examined prescription opioid use in a nested case-control study among community-based adults age 65 years or older in the state of Washington (1,039 cases and 2,022 controls) (34). That study found a significant association between prescription opioid use and the risk of community-acquired pneumonia with an odds ratio (OR) of 1.38 (95% CI: 1.08–1.76). In comparison, in a subgroup analysis, we observed an IRR of 1.22 (95% CI: 0.99, 1.51) for pneumonia alone and an IRR of 1.39 (95% CI: 1.19, 1.62) for all infections when comparing current use to non-use. That same study also found an increased OR for pneumonia of 3.24 (95% CI: 1.64, 6.39) comparing new use to non-use, which is similar to our increased IRR for all serious infections of 2.38 (95% CI: 1.65, 3.42) related to new use. Similarly, both studies observed an increased risk for infection related to immunosuppressive and long-acting opioid use. This similarity of findings across different study designs and populations provides further support to the observed association between opioid use and immunosuppression.

Several limitations in our study should be considered. Although our analyses accounted for both fixed and time-varying confounders, the possibility of residual confounding by indication cannot be excluded. The potential exists that within the population of patients with RA some unmeasured confounder that leads to current prescription opioid use also leads to an increased susceptibility to serious infection. One possibility would be that in general, pain leads to opioid analgesic use and an increased risk of infection. However, we found no significant association between NSAID use and the risk of infections, which argues against an obvious pain-related confounder affecting our results.

Although the TennCare computerized pharmacy records are an excellent source of medication data because they are not subject to information bias (46) and have high concordance with self-report of medication use (47–49), the possibility exists that patients did not take their opioid medications as prescribed (50). However, the within-person comparison nature of the SCCS method should at least in part account for an individual’s propensity to adhere to their prescriptions. Furthermore, although we tried to minimize potential exposure misclassification by introducing a recent use category, residual misclassification in our opioid exposure variable remains a possibility. As healthcare-associated infections were specifically excluded as an outcome by excluding all follow-up time during any hospitalization and in the 30-day post discharge period, our findings may not be generalizable to inpatient opioid use and infections that develop as the result of exposure to the healthcare setting. Finally, as only patients with RA enrolled in TennCare were considered in our study, our results may not be generalizable to other high-risk populations as well.

A strength of our study was the novel use of the SCCS method to examine the association between opioid use and serious infection while controlling for both fixed and time-varying covariates. If assumptions are satisfied, this strategy can provide an unbiased estimate of the association between opioid use and the risk of serious infection (31, 32). In addition, the consistency in the observed association between current opioid use compared to non-use across several planned secondary and sensitivity analyses provides evidence for the robustness of our study findings.

Our findings provide evidence of an association between the use of different opioid types and infection, but further research is needed to characterize the role of individual opioids. This remains an important question in understanding the degree to which previously identified biological mechanisms of immunosuppression for morphine, methadone, fentanyl and codeine might translate into clinically meaningful risks of infection in human populations. This is especially important given the current uncertainty about the long-term safety and effectiveness of opioid analgesics, especially among patients with RA and other vulnerable populations (20, 22).

Although our study lends support to the idea that opioids might cause further immunosuppression in patients with RA, these patients are already at higher risk of infection and possibly more likely to receive opioid therapy compared to other patient populations. Further studies are needed to determine whether this association exists in other patient populations.

In summary, in a population of adult patients with RA, opioid use was associated with an increased risk of hospitalizations due to serious infections. In addition, the incidence rate of infections was higher among periods of current opioid use at higher daily doses, when using long-acting formulations, and based upon the potential immunosuppressive properties of the opioid.