Abstract
Purpose
Longer duration of prescription opioid use is associated with risk of major depression after controlling for daily morphine equivalent dose and pain. It is not known if risk of depression varies as a function of the type of opioid prescribed.
Methods
A retrospective cohort design was used to model onset of new depression diagnosis among 11,462 Veterans Health Administration (VA) patients who were prescribed only codeine, only hydrocodone or only oxycodone for >30 days. Patients were free of prevalent opioid use and depression at baseline (2000-2001). Follow-up was 2002-2012. Propensity scores and weighting were used to balance covariates across opioid type. Cox-proportional hazard models were computed, using weighted data and additional adjustment for morphine equivalent dose (MED), duration of use, and pain after opioid initiation, to estimate the risk of new depression diagnosis among patients prescribed only codeine, only oxycodone vs. those prescribed only hydrocodone.
Results
After controlling for confounding, we observed that patients prescribed codeine, compared to hydrocodone, were significantly more likely to have a new depression diagnosis (HR=1.27; 95%CI: 1.12-1.43). Oxycodone was significantly associated with onset of new depression diagnosis when exposure was modeled as total days exposed in post-hoc analysis, but not when exposure was duration of incident period of use.
Conclusions
Although codeine is a less potent opioid, after controlling for MED, chronic use of this agent is associated with nearly a 30% greater risk of depression compared to hydrocodone. Additional research is needed to determine the mechanisms for this association.
Keywords: opioids, depression, retrospective cohort, epidemiology
INTRODUCTION
Several lines of evidence support the conclusion that chronic, greater than 90 days, and sub-chronic, greater than 30 day prescription opioid analgesic use increases the risk of new onset depression.1-3 The risk is independent of pain and daily morphine equivalent dose. To date, it is not known if the risk of new depression diagnosis among patients with no recent history of diagnosed depression differs between commonly prescribed opioids taken for 30 days or longer.
Existing studies of the correlates and consequences of specific opioid medications have generated mixed results. Oxycodone is related to subjects reporting more subjective effects, regardless of whether they are positive or adverse effects, compared to hydrocodone at typical prescription doses.4 In patients seeking opioid dependence treatment there were no significant differences in average Beck Depression Inventory scores in primary hydrocodone users compared to immediate release (IM) and extended release (ER) oxycodone users.5 Co-occurring illicit opioid use may be more common in patients using oxycodone in that significantly more oxycodone users reported ever using heroin than did hydrocodone users (32.9% ER oxycodone vs. 26.2% IR oxycodone vs. 16.3% hydrocodone).5 Several experimental abuse liability studies have found little evidence of difference in abuse potential between oxycodone and hydrocodone.6,7 Medications containing codeine and oxycodone do not differ in pain relief among patients with moderate to severe osteoarthritis8 and comparison of codeine and hydrocodone in treatment of chronic cancer pain found no differences in efficacy.9
A pharmacovigilence study of five opioid medications in non-cancer pain patients found greater risk of all-cause mortality after 30 days of use among codeine and oxycodone users compared to 30 days of hydrocodone use.10 In the same study, codeine, compared to hydrocodone, was significantly associated with more cardiovascular events at 180 days following medication initiation (RR=1.62; 95%CI:1.27-2.06). The increased risks associated with codeine and oxycodone were detected after matching patients on numerous covariates that could contribute to these outcomes.
It is not known whether specific opioid analgesics differ in contributing to risk of new depression diagnosis. In a large retrospective cohort design, controlling for pain and adjusting for morphine equivalent dose (MED) and duration of use, we sought to determine whether the hazard of new depression diagnosis differs among Veterans Health Administration (VA) patients prescribed only codeine, only hydrocodone or only oxycodone for 30 days or more.
METHODS
VA electronic medical record data, including ICD–9–CM diagnosis codes, prescription records, vital signs and demographic information, were used in analysis. The source file was a random sample of 500,000 VA patients, age 18-80, that used the VA between 2000-2012. These veterans represent service eras from World War II to present conflicts in Iraq and Afghanistan.
Cohort eligibility
We have established that continuous opioid use for 31-90 days and for > 90 days are both associated with increased risk of new depression diagnosis, compared to use limited to 1-30 days,1,3 therefore we restricted the sample to 11,462 VA patients with at least 30 days of use. Each type of opioid was thus prescribed for the minimum length of time associated with risk of new depression diagnosis. The follow-up period was 2002-2012. Follow-up continued until the onset of depression diagnosis or last available encounter. The study design included two years (2000-2001) for washout in which patients with a diagnosis of depression or any opioid use were excluded. Figure 1 shows the steps used in creating the analytic cohort.
Figure 1.
Cohort eligibility
Measures
Follow-up time
Observation or follow-up time was defined as months followed from Jan 1, 2002 to either depression diagnosis date or right censoring. Right-censoring time is month of last known visit relative to start of follow-up and is defined by individuals who do not meet criteria for depression before the end of follow-up. Cohort follow-up time was 2002 to 2012 while exposure follow-up time was from the start of the opioid prescription to either depression data or right censoring.
Opioid medication
The three most common opioids prescribed to VA patients for non-cancer pain were codeine, hydrocodone and oxycodone. Patients were free of opioids at baseline. During follow-up patients initiated either codeine, hydrocodone or oxycodone at any point prior to depression. We required a minimum of 30 days of continuous use of the initial opioid and did not allow for dual opioid use or switching. Duration of incident use was computed using the “days supply” variable and patients were assumed to be continuous users if there was no gap of 30 days or more between fills. Thus all patients used one of the three opioid medications for at least 30 days.
VA patient data comes from electronic medical records (EMR) and is not identical to insurance claims data because the medication record is recorded in the EMR at the time of fill. The VA has had a nationwide EMR since 1999 and therefore it would be difficult for a veteran to obtain opioids from more than one source within the VA without the pharmacy noticing and blocking the attempt. Every prescription fill is recorded in the EMR at the time of fill. We used the dates of initial fill, days of available pills, dates of refill and assumed the maximum daily dose was taken.
Outcome
Depression was defined by two or more outpatient diagnoses (ICD-9-CM codes = 296.2, 296.3 and 311) within the same 12 month period or at least one inpatient diagnosis for depression. This diagnostic algorithm has a 99% positive predictive value when chart review is the gold standard,11 and an 88% positive predictive value and 71% negative predictive value when self-reported lifetime depression is the gold standard.12 All patients were free of diagnosed depression for two years prior to baseline. During follow-up, patients who developed depression before opioid initiation were excluded because they did not provide informative data. The date of new onset depression during follow-up was defined as the date of a new hospitalization for depression or the date of the first of the two required ICD-9 diagnosis for depression in the same 12 month period.
Covariates
We modeled the effect of longer duration of use, defined as 31-90 days and >90 days, as a covariate. Covariates included morphine equivalent dose (MED) of each prescription. The last daily MED during the continuous use period was modeled as a binary variable of >50 mg vs. <50 mg MED. Anxiety disorders, including PTSD were defined by ICD-9-CM codes. Substance use disorders included nicotine dependence/history of smoking, alcohol and illicit drug abuse/dependence, the latter inclusive of opioid abuse/dependence. Co-medications included any benzodiazepine and any antidepressant prescribed for any dose and duration. Comorbid conditions included type 2 diabetes, obesity, low testosterone, sleep apnea and the Romano adapted Charlson Comorbidity Index.13 The Comorbidity Index is computed from the presence of 17 conditions associated with morbidity and mortality, and higher scores indicate worse health.13 Pain diagnoses were measured by ICD-9-CM codes for arthritis, back pain, headache, musculoskeletal pain and neuropathic pain. Pain score is available as a stand-alone clinical instrument in the VA and ranges from 0 to 10, with higher scores indicating worse pain. Pain score was defined and used in two separate ways: 1) maximum pain score before the end of follow-up to be used in propensity score weighting and 2) average monthly score in follow-up to be used in hazard models. If there was a month where pain score was not available, a score was carried forward until next available average monthly pain score. Thus, for each month of follow-up, a patient had a pain score available and we allowed for changing pain during follow-up. To control for detection bias that may be related to detecting new depression diagnosis at a higher rate among patients who use more health services, we modeled volume of clinic visits defined by quartiles of the average number of visits per month. Demographics, measured at baseline, included age, gender, race and marital status. Access to VA healthcare only vs. VA plus other coverage served as proxy for income and controlled for potential lower disease prevalence and greater chance of missing data in those with access to non-VA sector care
Propensity Scores and data weighting
Bias by indication, i.e. opioid prescribing is associated with pain that in turn in associated with new depression diagnosis, may confound the association between opioid medication and new depression diagnosis as it is observed in clinical practice. To ensure that Cox proportional survival models measured the association between type of opioid and new diagnosis of depression not confounded by factors predicting exposure and new depression diagnosis, 14-17 we sought to equate the distribution of covariates among patients prescribed only codeine, only hydrocodone or only oxycodone using propensity scores (PS) and inverse probability of treatment weighting (IPTW). The PS is a conditional probability that a patient will receive 30 or more days of a specific opioid. The propensity score was computed using multivariable, multinomial logistic regression, predicting exposure to a specific opioid given covariates listed in Table 1, except for opioid analgesic use duration and MED. We did not include duration and dose of opioid analgesic use in PS models because these variables are measured after the initiation of the opioid. After obtaining a PS, we calculated stabilized weights.18 The stabilized weight is the marginal probability of exposure, or exposure to a specific opioid given no covariates, divided by the PS. Weighting results in pseudopopulations for each exposure group such that covariates listed in Table 1 balance across the exposure.14-16 In the weighted analyses, covariates are balanced and survival models are not confounded by factors that predict type of opioid received. Propensity models perform well if covariates achieve balance across groups and stabilized weights have a mean near one and don’t fall outside a maximum of 10.19 The IPTW stabilized weights in this study ranged from 0.43 to 3.02, with a mean of 1.00 (sd=0.17) and median of 0.97.
Table 1. Association of covariates and incident depression with type of opioid, > 30 day users (2002-2012) (n=11,462).
| Outcome/covariates, n(%) | Overall (n=11,462) |
Codeine (n=2,622) |
Hydrocodone (n=7,225) |
Oxycodone (n=1,615) |
p-value |
|---|---|---|---|---|---|
| Outcomes | |||||
| Depression, cumulative incidence (2002-2012) |
1459 (12.7) | 379 (14.5) | 841 (11.6) | 239 (14.8) | <.0001 |
| Depression, incidence rate (PY=person-years) |
14.3/1000 PY | 16.9/1000 PY | 12.7/1000 PY | 17.2/1000 PY | <.0001 |
| Follow-up time (months), mean (sd) | 107.6 (34.1) | 103.1 (36.8) | 110.2 (32.1) | 103.7 (37.2) | <.0001 |
| Follow-up time (months), median | 129.0 | 127.0 | 129.0 | 128.0 | <.0001 |
| Covariates | |||||
| Opioid duration (days) | <.0001 | ||||
| 31-90 | 6667 (58.2) | 1672 (63.8) | 3882 (53.7) | 1113 (68.9) | |
| > 90 | 4795 (41.8) | 950 (36.2) | 3343 (46.3) | 502 (31.1) | |
| Last Daily Dose: > 50 mg | 332 (2.9) | 8 (0.3) | 164 (2.3) | 160 (9.9) | <.0001 |
| Benzodiazepine use | 3665 (32.0) | 768 (29.3) | 2299 (31.8) | 598 (37.0) | <.0001 |
| Antidepressant use | 7279 (63.5) | 1650 (62.9) | 4549 (63.0) | 1080 (66.9) | .010 |
| Psychiatric comorbidities # | |||||
| PTSD | 1745 (15.2) | 375 (14.3) | 1077 (14.9) | 293 (18.1) | .002 |
| Other anxiety * | 1538 (13.4) | 333 (12.7) | 984 (13.6) | 221 (13.7) | .469 |
| Nicotine abuse/dependence | 5115 (44.6) | 1125 (42.9) | 3224 (44.6) | 766 (47.4) | .016 |
| Alcohol abuse/dependence | 2273 (19.8) | 477 (18.2) | 1429 (19.8) | 367 (22.7) | .002 |
| Any illicit drug abuse/dependence | 1346 (11.7) | 300 (11.4) | 800 (11.1) | 246 (15.2) | <.0001 |
| Comorbidities # | |||||
| Diabetes Type II | 5070 (44.2) | 1164 (44.4) | 3192 (44.2) | 714 (44.2) | .982 |
| Obesity diagnosis | 4495 (39.2) | 1007 (38.4) | 2823 (39.1) | 665 (41.2) | .184 |
| Low T | 359 (3.1) | 75 (2.9) | 224 (3.1) | 60 (3.7) | .291 |
| Sleep apnea | 1202 (10.5) | 277 (10.6) | 742 (10.3) | 183 (11.3) | .448 |
| Comorbidity index, mean (sd) | 2.9 (2.7) | 3.1 (2.9) | 2.8 (2.7) | 3.0 (2.8) | <.0001 |
| Painful conditions # | |||||
| Arthritis | 9806 (85.6) | 2283 (87.1) | 6169 (85.4) | 1354 (83.8) | .012 |
| Back pain | 8128 (70.9) | 1845 (70.4) | 5141 (71.2) | 1142 (70.7) | .734 |
| Headaches | 2206 (19.3) | 555 (21.2) | 1331 (18.4) | 320 (19.8) | .008 |
| Musculoskeletal pain | 7057 (61.6) | 1626 (62.0) | 4384 (60.7) | 1047 (64.8) | .007 |
| Neuropathic pain | 4044 (35.3) | 975 (37.2) | 2471 (34.2) | 598 (37.0) | .007 |
| Maximum pain score, mean (sd) | 8.6 (2.0) | 8.6 (2.0) | 8.5 (2.0) | 8.9 (1.8) | <.0001 |
| High Healthcare utilization (top 25%) | 2996 (26.1) | 817 (31.2) | 1638 (22.7) | 541 (33.5) | <.0001 |
| Age | 56.7 (12.7) | 58.0 (13.0) | 56.5 (12.7) | 55.2 (12.1) | <.0001 |
| Gender: male | 10898 (95.1) | 2496 (95.2) | 6872 (95.1) | 1530 (94.7) | 0.780 |
| Race: White | 9220 (80.4) | 2006 (76.5) | 5939 (82.2) | 1275 (78.9) | <.0001 |
| Insurance: VA only | 7208 (62.9) | 1574 (60.0) | 4553 (63.0) | 1081 (66.9) | <.0001 |
| Marital status: Married | 6460 (56.4) | 1451 (55.3) | 4155 (57.5) | 854 (52.9) | .002 |
Comorbidities occurring before incident depression
Other anxiety disorders = panic disorder, OCD, social phobia, GAD, Anxiety NOS
Analysis
All analyses were performed using SAS v9.4 (SAS Institute, Cary, NC). Weighted analyses included a ‘weight’ statement in SAS procedures to apply stabilized IPTW. Unweighted, bivariate analyses assessed the relationship of covariates with opioid type using ANOVA for continuous variables and chi–square tests for categorical variables. After IPTW, bivariate analyses were re-computed to determine that covariates balanced across type of opioid. A chi-square test and Poisson regression model were used to compare unweighted cumulative incidence and incidence rate (person-years) during 2002-2012 across opioid type. Mean and median follow-up time were compared between groups using an ANOVA and non-parametric Kruskal-Wallis test, respectively. Hazard ratios for new depression diagnosis were estimated using Cox proportional hazards models in which the exposure, i.e. type of 30 day incident opioid use was treated as a time dependent variable. All other variables except health care utilization and demographics were time dependent. Time dependent variables allow for using all available information for a patient to ascertain when status changes from unexposed to exposed (e.g. a new diagnosis of anxiety or any other comorbidity) any time before depression or last available data point. Thus, exposure time isn’t necessarily time since baseline. For instance, a subject’s time spent exposed to an opioid was calculated from the date of initiating a prescription, whether that was the date at baseline or at some time during follow-up. Also, variables used in the propensity score model could occur any time before incident depression, including during the period of opioid use. Follow-up time was in months and ended when depression diagnosis occurred or at the end of last available data.
Separate Cox proportional hazard models were computed prior to weighting, after weighting, and after weighting and further adjustment for duration of use (i.e. 30-90 days and >90 days) and MED, and last, a full model was fit that included weighted data, control for duration of use, MED, painful conditions and self-reported pain score. The common reference group in all survival models was hydrocodone because it was the most common incident opioid prescribed in this patient sample. The PHREG procedure in SAS version 9.4 (SAS Institute, Cary, NC) with α set at 0.05 was used for the Cox regression models. Evaluation of hazard trends over time confirmed that proportional hazard assumptions were met (p=0.10). Two-tailed tests were conducted to allow for both risk factors and protective effects. This project was approved by the Institutional Review Boards of participating institutions.
RESULTS
Among the eligible patients, 22.9% were prescribed only codeine, 63.0% prescribed only hydrocodone and 14.1% only oxycodone for at least 30 days. Overall incidence rate of new depression diagnosis across 2002 to 2012 was 14.3/1000 person-years. New diagnoses of depression differed across groups such that codeine and oxycodone were similar at around 17/1000 person-years followed by hydrocodone at 12.7/1000 person-years. The distribution of covariates by specific opioids is shown in Table 1. Opioid use of 30-90 days was most common among oxycodone users and opioid use > 90 days was most common among hydrocodone users (p<0.0001). Having a daily dose before end of follow-up >50 mg MED was most common among oxycodone users (p<0.0001). PTSD and substance use disorders, including nicotine dependence, were significantly more common among patients receiving oxycodone (p<0.05 - p<0.0001). Benzodiazepine (p<0.0001) and antidepressant (p=0.010) co-medications were most common among oxycodone users.
The distribution of individual comorbid conditions did not significantly differ across the three types of opioids. However, the comorbidity index measure was largest among codeine using patients (p<0.0001). Significant associations were observed between the prevalence of painful conditions and type of opioid prescription. Arthritis and headache were significantly more common among patients prescribed only codeine (p=0.012 and p=0.008), and musculoskeletal pain diagnosis was significantly more common among patients prescribed only oxycodone (p=0.007). Neuropathic pain was more common among those prescribed only codeine and only oxycodone (p=0.007) and the average maximum pain score was largest among patients prescribed only oxycodone (p<0.0001). There was no significant association between back pain and type of opioid prescription.
A significantly higher volume of health care utilization was observed in patients prescribed only codeine and only oxycodone (p<0.0001). Older age (p<0.0001) was associated with prescription for only codeine, and white race (p<0.0001) and being currently married (p=0.002) was most common among patients prescribed only hydrocodone. Access to VA healthcare only was most common among those prescribed only oxycodone.
After applying stabilized weights derived from IPTW, the association between covariates and type of opioid use was recomputed and is shown in Table 2. For all covariates, weighting was successful in removing the significant differences in the distribution of covariates by type of opioid and generated prevalence estimates that were very similar across each type of opioid.
Table 2. Association of covariates with opioid type, weighted by inverse probability of exposure, > 30 day users(2002-2012) (n=11,462).
| Covariates, % | Codeine (n=2,622) |
Hydrocodone (n=7,225) |
Oxycodone (n=1,615) |
p-value |
|---|---|---|---|---|
| Benzodiazepine use | 32.0 | 32.0 | 32.1 | .998 |
| Antidepressant use | 63.7 | 63.5 | 63.2 | .951 |
| Psychiatric comorbidities # | ||||
| PTSD | 15.3 | 15.2 | 15.2 | .989 |
| Other anxiety * | 13.7 | 13.4 | 13.3 | .920 |
| Nicotine abuse/dependence | 44.5 | 44.6 | 44.3 | .958 |
| Alcohol abuse/dependence | 20.0 | 19.9 | 19.9 | .988 |
| Any illicit drug abuse/dependence |
11.6 | 11.7 | 11.7 | .985 |
| Comorbidities # | ||||
| Diabetes Type II | 44.7 | 44.3 | 44.6 | .926 |
| Obesity diagnosis | 39.3 | 39.2 | 39.2 | .993 |
| Low T | 3.1 | 3.1 | 3.1 | .991 |
| Sleep apnea | 10.7 | 10.5 | 10.4 | .947 |
| Comorbidity index, mean (sd) | 2.9 (2.8) | 2.9 (2.7) | 2.9 (2.7) | .966 |
| Painful conditions # | ||||
| Arthritis | 85.3 | 85.5 | 85.3 | .953 |
| Back pain | 70.7 | 70.9 | 71.3 | .920 |
| Headaches | 19.3 | 19.3 | 19.2 | .998 |
| Musculoskeletal pain | 61.6 | 61.6 | 60.9 | .865 |
| Neuropathic pain | 35.4 | 35.3 | 34.7 | .905 |
| Maximum pain score, mean (sd) | 8.6 (2.0) | 8.6 (2.0) | 8.5 (2.1) | .928 |
| High Healthcare utilization (top 25%) |
26.2 | 26.1 | 26.2 | .994 |
| Age | 56.6 (13.0) | 46.7 (12.7) | 56.6 (12.3) | .977 |
| Gender: male | 94.9 | 95.0 | 95.1 | .961 |
| Race: White | 80.4 | 80.5 | 80.4 | .990 |
| Insurance: VA only | 62.7 | 62.8 | 62.7 | .992 |
| Marital status: Married | 56.2 | 56.3 | 56.2 | .996 |
Comorbidities occurring before incident depression
Other anxiety disorders = panic disorder, OCD, social phobia, GAD, Anxiety NOS
Cox proportional hazard modeling results are shown in Table 3. Prior to IPTW in model 1, compared to patients prescribed only hydrocodone for 30 days or longer, patients prescribed only codeine or only oxycodone for 30 days or longer, had a significantly greater risk for new diagnosis of depression, (HR=1.35; 95%CI:1.20-1.52 and 1.37; 95%CI: 1.19-1.58, respectively). In weighted data, model 2, codeine remained significantly associated with greater risk of new depression diagnosis and oxycodone was no longer significantly associated with new depression diagnosis as compared to hydrocodone. Hazard ratios remained similar after adjusting for duration and dose in model 3. After additional control for pain that could occur after initiation of an opioid and for changing level of pain across follow-up time in model 4, patients prescribed only codeine were significantly more likely to develop a new diagnosis of depression compared to patients prescribed only hydrocodone (HR=1.27; 95%CI:1.12-1.43). Patients prescribed only oxycodone were not at a significantly increased risk of new depression diagnosis compared to those only prescribed hydrocodone (HR=1.11; 95%CI:0.95-1.29). In this full model, opioid MED and duration were not significantly associated with new depression diagnosis. Back pain, headache, musculoskeletal pain, neuropathy and pain scores were all significantly associated with increased risk of new diagnosis of depression (HR range:1.08-1.64).
Table 3. Association between opioid drug type for incident use and incident depression, unweighted and weighted by inverse probability of exposure, > 30 day users (n=11,462).
| Model 1 − Crude a |
Model 2− Weighted b |
Model 3− Weighted + Dose+ Duration c |
Model4 − Weighted + Pain d |
|
|---|---|---|---|---|
| Variable | HR (95% CI) | HR (95% CI) | HR (95% CI) | HR (95% CI) |
| Opioid type | ||||
| Codeine | 1.35 (1.20-1.52) | 1.31 (1.16-1.47) | 1.30 (1.15-1.46) | 1.27 (1.12-1.43) |
| Hydrocodone | 1.00 | 1.00 | 1.00 | 1.00 |
| Oxycodone | 1.37 (1.19-1.58) | 1.15 (0.99-1.33) | 1.14 (0.98-1.34) | 1.11 (0.95-1.29) |
| Opioid duration (days) | ||||
| 31-90 | 1.00 | 1.00 | ||
| > 90 | 0.96 (0.86-1.07) | 0.94 (0.85-1.05) | ||
| Last Daily Dose: > 50 mg | 0.93 (0.68-1.29) | 0.88 (0.63-1.21) | ||
| Arthritis | 1.05 (0.91-1.22) | |||
| Back pain | 1.56 (1.38-1.77) | |||
| Headache | 1.64 (1.46-1.85) | |||
| Musculoskeletal pain | 1.55 (1.38-1.74) | |||
| Neuropathy | 1.18 (1.05-1.32) | |||
| Pain Score | 1.08 (1.06-1.10) |
HR=hazard ratio; CI=confidence interval
p=.854 for comparing oxycodone and codeine
p=.135 for comparing oxycodone and codeine
p=.154 for comparing oxycodone and codeine
p=.126 for comparing oxycodone and codeine
DISCUSSION
In a large VHA patient sample, we compared the hazard of new depression diagnosis in patients that were prescribed only codeine, only hydrocodone or only oxycodone, for 30 days or longer. Patients prescribed only codeine for 30 days or longer had a 29% increased risk of new diagnosis of depression compared to those prescribed only hydrocodone for 30 days or longer. Those prescribed only oxycodone for 30 days or longer were not significantly more likely to develop new depression diagnosis compared to patients prescribed only hydrocodone for 30 days or more. This effect remained after controlling for confounding and bias by indication using PS and IPTW and after additional adjustment for duration of opioid use, MED and for the contribution of pain after initiation of opioid therapy.
We have previously demonstrated3 in three separate patient samples with wide ranging demographic differences, that patients who use opioids for 30-90 days have an 18% to 33% greater risk of new depression diagnosis compared to patients who are 1-30 day users. Patients who use opioids for longer than 90 days had between 40% to 130% greater risk of new diagnosis of depression compared to patients whose opioid analgesic use did not exceed 30 days. The present study suggests this increased risk is greater in patients only receiving codeine compared to patients prescribed only hydrocodone. Our observation that risk of depression is greater among patients receiving codeine was unexpected because we anticipated no differences would occur given the literature that opioids do not differ on other outcomes such as abuse potential5-7 and effective pain relief.8,9
Other studies have found an increased rate of adverse events with codeine use in older patients initiating opioids for non-cancer pain. After 180 days of follow-up, Solomon et al.10 determined that codeine, compared to hydrocodone, was significantly associated with greater risk of cardiovascular events, hospitalization for safety event, hospitalized death and all-cause mortality. With the exception of hospitalization for safety event, these authors also found oxycodone, compared to hydrocodone was associated with a similar magnitude of adverse events. However it is not known why codeine imparted greater risk for adverse events when compared to hydrocodone. One possibility is ineffective pain relief among those patients, approximately 10% of the Caucasian population, is unable to convert codeine to morphine.20,21 . Several types of antidepressants including SSRIs and bupropion can reduce the analgesic effects of codeine via inhibition of its conversion to morphine.22 In this case, patients with depression symptoms taking an SSRI may experience less pain relief. However, we adjusted for pain scores following initiation of opioids. Maybe the lack of pain relief and not severity of pain leads to patient frustration and helplessness that might contribute to depression.
Limitations of the present study include lack of data on adherence. We are unable to confirm whether opioids were taken as prescribed. We did not account for opioid prescriptions that might have been written between the end of the incident period of use and the end of follow-up. Thus patients could have started a different opioid after a 30 day gap and prior to the onset of a new depression diagnosis. To determine if our criteria for opioid exposure resulted in mis-classifying which patients were exposed to codeine, hydrocodone and oxycodone, we tested an alternative opioid exposure variable by dropping the continuous use criterion and adding up all days on which an opioid was available to the patient before onset of depression diagnosis. This approach demonstrated an association with new diagnosis of depression among patients only exposed to codeine (HR=1.52; 95%CI:1.43-1.62) and only exposed to oxycodone (HR=1.36; 95%CI:1.26-1.47), furthering support for a lower risk of new diagnosis of depression among hydrocodone only users.
The validity of depression diagnosis11,12 does not necessarily apply to incident depression. Therefore it is possible that patients who developed depression have sub-clinical symptoms made worse by opioids or have a history of depression not recorded in the medical record. Given routine screening for depression in the VHA, it seems unlikely that we misclassified cases as non-cases. It is possible that depression symptoms and lifetime history, both contributors to risk of further depression episodes were present in both patients who did and did not develop depression during our follow-up period.
Results may not generalize to veterans who do not use the VA health care system. VA patient populations, compared to civilian populations, have more physical and psychiatric comorbidity and tend to have fewer socioeconomic resources. The higher than national average prevalence of PTSD and other psychopathology likely increases the risk of depression which may in turn reduce the ability to detect the association between opioid use and depression.
Previously reported patterns of opioid prescribing23 could impact our results. Over half of the new opioid prescriptions in our analysis occurred between 2002 and 2005 during which codeine decreased from 19 to 14%, hydrocodone increased from 22 to 30% and oxycodone decreased from 23.7 to 19.3%. If the decreasing level of exposure to codeine and hydrocodone in the VA resulted in lower hazard of depression related to these medications then our hazard estimates may be conservative.
Additional research is needed to determine if dose escalation differs between these medications and whether trajectories of dose escalation, both rapid increases and decreases would make patients more prone to depression. Extended release medications were not excluded. However, sensitivity analysis removing the 11 patients initiated on extended release oxycodone did not change results. In our models, patients were users of only one type of opioid, therefore results are not be applicable to patients who, for example, initiate codeine and then switch to another opioid or have another opioid added to their codeine regimen. Last, results may not generalize to non-VA patient populations with less comorbidity and a higher proportion of women. However, our previous studies of opioid use and incident depression 1,2,3 and depression relapse24 have produced very similar findings in two private sector patient samples.
Strengths of this study include the large sample size, control for confounding by indication and the high quality pharmacy data. VA pharmacy data is recorded at the time of prescription fill, and it is highly unlikely that the type of opioid was coded incorrectly. While we did not measure severity of depression, our on-going research indicates that the clinical characteristics of depression diagnosis following opioid use are similar to those of depression in non-opioid use. For instance, we found that 71.5% of patients with depression that onset after opioid use received an antidepressant which was not significantly different from the 69.5% of patients with depression who did not use opioids. Nearly identical percentages were observed for receipt of acute phase, 12 week antidepressant treatment. In the opioid related depression patients, 51.5% received acute phase treatment compared to 50.3% of non-opioid users with depression. Thus the disease severity of depression in our study is likely the same as new onset depression in general.
There are several potential biologic mechanisms for the opioid, new onset depression association. Chronic OAU may cause depression symptoms via androgen deficiency which occurs in 75% of chronic opioid using men and women.25 Others have reported persistent neurophysiological changes in the nucleus accumbens and amygdala following long-term opioid exposure.26,27 Because both brain regions are associated with mood, reward and motivation, this might be the biological mechanism for our findings.
We speculate there are potential clinical implications that need to be confirmed with prospective studies. First, clinicians should be aware that risk of a new depression diagnosis with > 30 day opioid use is greater in patients prescribed only codeine compared to those prescribed only hydrocodone. Patients using any of these opioids, and especially those prescribed codeine, should be routinely screened for depression. Providers should discuss risk of depression with patients before initiating chronic opioid therapy and when considering the relative risks of specific medications.
Key points.
Controlling for dose, > 30 day use of prescription codeine only was associated with increased hazard of new depression diagnosis compared to >30 day use of hydrocodone only.
Prescription of oxycodone only was not associated with increased onset of new depression diagnosis in planned models using duration of incident period of use. In post-hoc analysis, when oxycodone use was measured by total days of exposure, it was significantly associated with depression compared to hydrocodone total days exposed.
We speculate that patients and clinicians should consider the elevated risk of new onset depression that may be associated with codeine and oxycodone.
Acknowledgments
Funding Support: This study was supported by the National Institute of Mental Health, Prescription Opioid Analgesics and Risk of Depression, R21MH101389. Funding sources had no role in study design; in the collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit the paper for publication.
Footnotes
Conflict of interest: none
Prior postings and presentations: none
REFERENCES
- 1.Scherrer JF, Svrakic DM, Freedland KE, et al. Prescription opioid analgesics increase the risk of depression. J Gen Intern Med. 2014;29(3):491–499. doi: 10.1007/s11606-013-2648-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Scherrer JF, Salas J, Lustman PJ, Burge S, Schneider FD, Texas RRNo Change in opioid dose and change in depression in a longitudinal primary care patient cohort. Pain. 2015;156(2):348–355. doi: 10.1097/01.j.pain.0000460316.58110.a0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Scherrer JF, Salas J, Copeland LA, et al. Prescription opioid duration, dose, and increased risk of depression in 3 large patient populations. Ann Fam Med. 2016;14:54–62. doi: 10.1370/afm.1885. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Zacny P, Gutierrez S. Subjective, psychomotor, and physiological effects profile of hydrocodone/acetaminophen and oxycodone/acetaminophen combination products. Pain Medicine. 2008;9:433–443. doi: 10.1111/j.1526-4637.2007.00359.x. [DOI] [PubMed] [Google Scholar]
- 5.Nielsen S, Hillhouse M, Weiss RD, et al. The relationship between primary prescription opioid and buprenorphine-naloxone induction outcomes in a prescription opioid dependent sample. Am J Addict. 2014;23:343–348. doi: 10.1111/j.1521-0391.2013.12105.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Zacny JP, Gutierrez S. Within-subject comparison of the psychopharmacological profiles of oral hydrocodone and oxycodone combination products in non-drug-abusing volunteers. Drug Alcohol Depend. 2009;101:107–114. doi: 10.1016/j.drugalcdep.2008.11.013. [DOI] [PubMed] [Google Scholar]
- 7.Walsh SL, Nuzzo PA, Lofwall MR, HJ JR. The relative abuse liability of oral oxycodone, hydrocodone and hydromorphone assessed in prescription opioid abusers. Drug Alcohol Depend. 2008;98:191–202. doi: 10.1016/j.drugalcdep.2008.05.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Corsinovi L, Martinelli E, Fonte G, et al. Efficacy of oxycodone/acetaminophen and codeine/acetaminophen vs. conventional therapy in elderly women with persistent, moderate to severe osteoarthritis-related pain. Arch Gerontol Geriatr. 2009;49:378–382. doi: 10.1016/j.archger.2008.12.003. [DOI] [PubMed] [Google Scholar]
- 9.Rodriguez RF, Castillo JM, Cstillo MDP, et al. Codeine/acetaminophen and hydrocodone/acetaminophen combination tablets for the management of chronic cancer pain in adults: A 23-day, prospective, double-blind, randomized, parallel-group study. Clinical Therapeutics. 2007;29:581–587. doi: 10.1016/j.clinthera.2007.04.004. [DOI] [PubMed] [Google Scholar]
- 10.Solomon DH, Rassen JA, Glynn RJ, Garneau K, Levin R, Schneeweiss S. The comparative safety of opioids for nonmalignant pain in older adults. Archives of Internal Medicine. 2010;170:1979–1986. doi: 10.1001/archinternmed.2010.450. [DOI] [PubMed] [Google Scholar]
- 11.Solberg LI, Engebretson KI, Sperl-Hillen JM, Hroscikoski MC, O’Connor PG. Are claims data accurate enough to identify patients for performance measures or quality improvement? The case of diabetes, heart disease and depression. American Journal of Medical Quality. 2006;21:238–245. doi: 10.1177/1062860606288243. [DOI] [PubMed] [Google Scholar]
- 12.Frayne SM, Miller DR, Sharkansky EJ, et al. Using adminstrative data to identifymental illness: What approach is best? American Journal of Medical Quality. 2010;25(1):42–50. doi: 10.1177/1062860609346347. [DOI] [PubMed] [Google Scholar]
- 13.Romano PS, Roos LL, Jollis JG. Adapting a clinical comorbidity index for use with ICD9-CM administrative data: Differing perspectives. Journal of Clinical Epidemiology. 1993;46:1075–1079. doi: 10.1016/0895-4356(93)90103-8. [DOI] [PubMed] [Google Scholar]
- 14.Cole SR, H M. Constructing inverse probability weights for marginal structural models. American Journal of Epidemiology. 2008;168:656–664. doi: 10.1093/aje/kwn164. 2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Curtis LH, H B, Eisenstein EL, Kramer JM, Anstrom KJ. Using inverse probability-weighted estimators in comparative effectiveness analysis with observational databases. Medical Care. 2007;45:S103–S107. doi: 10.1097/MLR.0b013e31806518ac. [DOI] [PubMed] [Google Scholar]
- 16.Kilpatrick RD, G D, Brookhart MA, Polley E, Rothman KJ, Bradbury BD. Exploring large weight deletion and the ability to balance confounders when using inverse probability of treatment weighting in the presence of rate treatment decisions. Pharmacoepidemiol Drug Safety. 2013;22:111–121. doi: 10.1002/pds.3297. [DOI] [PubMed] [Google Scholar]
- 17.Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika Trust. 1983;70:41–55. [Google Scholar]
- 18.Xu S, Ross C, Raebel MA, Shetterly S, Blanchette C, Smith D. Use of stabilized inverse propensity scores as weights to directly estimate relative risk and its confidence intervals. Value Health. 2010;13:273–277. doi: 10.1111/j.1524-4733.2009.00671.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Sturmer T, Wyss R, Glynn RJ, Brookhart MA. Propensity scores for confounder adjustment when assessing the effects of medical interventions using nonexperimental designs. J. Intern. Med. 2014;275:570–580. doi: 10.1111/joim.12197. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Crews KR, Gaedigk A, Dunnenberger HM, et al. Clinical pharmacogenetics implementation consortium guidelines for cytochrome P450 2D6 genotype and codeine therapy: 2014 update. Clinical Pharmacology & Therapeutics. 2012;95:376–382. doi: 10.1038/clpt.2013.254. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Caraco Y, Sheller J, Wood AJ. Impact of ethnic origin and quinidine coadministration on codeine’s disposition and pharmacodynamic effects. J Pharmacol Exp Ther. 1999:413–422. [PubMed] [Google Scholar]
- 22.Vallejo R, Barkin RL, Wang VC. Pharmacology of opioids in the treatment of chronic pain syndromes. Pain Physician. 2011;14:E343–E360. [PubMed] [Google Scholar]
- 23.Bohnert ASB, Ilgen MA, Trafton JA, et al. Trends and regional variation in opioid overdose mortality among Veterans Health Administration patients, fiscal year 2001 to 2009. Clin. J. Pain. 2014;30:605–612. doi: 10.1097/AJP.0000000000000011. [DOI] [PubMed] [Google Scholar]
- 24.Scherrer JF, Salas J, Copeland LA, et al. Increased risk of depression recurrence after initiation of prescription opioids in non-cancer pain patients. J. Pain. doi: 10.1016/j.jpain.2015.12.012. In Press. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Smith HS, Elliott JA. Opioid-induced androgen deficiency (OPIAD) Pain Physician. 2012;15:ES145–ES156. [PubMed] [Google Scholar]
- 26.Upadhyay J, Maleki N, Potter J, et al. Alterations in brain structure and functional connectivity in prescription opioid-dependent patients. Brain. 2010;133:2098–2114. doi: 10.1093/brain/awq138. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Younger JW, Chu LF, D’Arcy N, Trott K, Jastrzab LE, Mackey SC. Prescription opioid analgesics rapidly change the human brain. Pain. 2011;152(8):1803–1810. doi: 10.1016/j.pain.2011.03.028. [DOI] [PMC free article] [PubMed] [Google Scholar]