Past studies and commentary have claimed that higher opioid analgesic doses increase mortality rates; yet, the research methods used to establish this should be challenged by critical readers. A recent data-mining investigation from Canada is a good example of how evidence can be statistically manipulated to arrive at outcomes with questionable validity, perpetuating myths of opioid harms that may not truly exist.
According to Tara Gomes, MHSc, and colleagues from Canada, writing in the Archives of Internal Medicine, opioids are often prescribed for chronic noncancer pain at doses exceeding those recommended in clinical practice guidelines and this incurs increased mortality [Gomes et al. 2011A]. To examine the risk-benefit ratio of such high-dose opioid therapy they developed a retrospective, population-based, case-control investigation, spanning more than a 9-year period (August 1, 1997, through December 31, 2006). Subjects came from a database of residents in Ontario, Canada, aged 15 to 64 years, who were eligible for publicly funded prescription drug coverage and had received an opioid analgesic for nonmalignant pain.
The primary outcome of interest was cases of opioid-related deaths, as determined by the investigating coroners, compared with control subjects, who were persons receiving comparable opioid analgesics during the 9+ years time period and did not die of opioid-related causes but were otherwise matched to the cases on a array of characteristics. The risk of opioid-related death was compared across patients treated with varying daily doses of opioids: 1-19 mg/day morphine or equivalent dose (MED) was the reference category, as compared with 20-49, 50-99, 100-199, or ≥200 mg/d MED.
Among 607,156 persons aged 15 to 64 years prescribed an opioid during the study period, the researchers identified 498 eligible case patients whose deaths were allegedly related to the opioids and 1,714 matched controls. After extensive adjustments for multiple clinical and demographic factors, they found that an average daily dose of ≥200 mg/d MED was associated with a nearly 3-fold increase in the “risk” of opioid-related mortality (Odds Ratio [OR]=2.88; 95% Confidence Interval [CI] 1.79-4.63) relative to low daily doses (<20 mg/d MED). They also report lesser but significant increases in opioid-related mortality with intermediate doses of opioids (50-99 mg/d MED, OR=1.92, 95% CI 1.30-2.85; and, 100-199 mg/d MED, OR=2.04, 95% CI 1.28-3.24 [also see figure and discussion below]).
The authors conclude that among patients receiving opioids for nonmalignant pain, the daily dose is strongly associated with opioid-related mortality, particularly at doses exceeding thresholds recommended in recent clinical guidelines (eg, ≥200 mg/d MED). They state, “We believe physicians should carefully assess the appropriateness of long-term use of opioids to treat chronic, non-cancer-related pain, particularly at high doses.”
COMMENTARY: The best that can be said of data-mining ventures such as this study is that they are sophisticated computerized simulations of what might be occurring in a particular population, that they are suggestive rather than confirmative of any cause-effect relationships, and that they have considerable limitations and sources of bias. Unfortunately, the results are too often widely accepted as fact, cited in the literature, and become part of a growing mythology of opioid risks, which we also recently discussed in an UPDATE [here].
This study is yet another example of how opioid harms have been “myth-represented” via the creative use of data mining and statistical prestidigitation. We will leave the most disconcerting observations regarding this study for the end (below) and first focus on some questionable evidentiary points.
- All subjects in the database were aged 15 to 64 years who qualified for the Ontario Public Drug Plan; this included patients on disability, residents in long-term care facilities or receiving home care services, and unemployed or economically disadvantaged persons. Therefore, this was a select population at the outset that may not typify the general public — hence, external validity is doubtful.
- Opioid-related mortality was derived from coroner determinations that “a combination of drugs (including ≥1 opioid) resulted in death, or those in which forensic toxicologic testing revealed opioid concentrations sufficiently high to cause death….” Such assessments have been disputed in past research, in that (a) the opioid(s) in question may be merely present as part of a mix of agents but not directly or solely causative of death, and (b) the postmortem concentration in serum or blood of any opioid that is judged as representing toxicity in a given person depends on multiple factors and is usually vaguely defined.
Case subjects in this study were taking greater percentages of antidepressants, other psychotropic agents or CNS depressants, and methadone than were control subjects. Alcohol use was unaccounted for, but about 85% of case subjects were using benzodiazepines compared with only 64% of controls, which may have influenced mortality when combined with opioids or alcohol. Furthermore, the authors included all types of opioid-related deaths in their analyses, including nearly 17% of which were suicide and probably unrelated to the prescribed opioid dose itself (unless the dose was so inadequate that decedents had intolerable pain).
- In lieu of an examination of individual patient charts, the researchers had to use complex and imprecise algorithms to determine opioid dosing. The authors state that they “estimated the average daily dose as the total quantity of opioids intended for use in the 120 days prior to the index date (in milligrams of morphine equivalents) divided by 120” [emphasis added]. There was no way of knowing the exact prescription subjects received, if they took the opioids as prescribed, or if they supplemented their prescriptions with additional opioids or other agents that went unrecorded in the database. Therefore, at most, we must assume that the researchers were only guessing at dosing regimens and patient compliance, and the accuracy is questionable.
- A comparison with <20 mg/d MED as a reference was probably inappropriate, since it is unlikely that persons with serious chronic pain conditions would benefit from such a low dose on a consistent basis. The authors do not indicate how many case and control subjects were in this category, but using such a low threshold as a reference point probably arbitrarily inflated the odds ratios in the other dose categories.
- Furthermore, in a just-published letter to the journal editor, Owen Williamson, MBBS, from Australia, looking closely at the data (see figure at right adapted from Gomes et al.), points out that the proportions of deaths (or event rates) for cases in the 20-49 mg and the ≥200 mg categories are identical (0.23), and the lower dose range (20-49 mg) — with its confidence interval crossing 1.0 — was not significantly different statistically from the reference range (<20 mg). Furthermore, the odds ratio confidence intervals overlap for all dose categories, suggesting that mortality is not associated with opioid dose alone [Williamson 2011]. [Concepts of confidence intervals were discussed in an UPDATE here.]
In a rebuttal letter, the researchers claim Williamson misinterpreted their study [Gomes et al. 2011B]. They assert that in data not published in their article “the association between dose and opioid-related mortality was consistent in our unadjusted conditional logistic regression model for the primary outcome….” Furthermore, while “the 95% confidence intervals overlap between high and moderate dose categories, a clear and significant association between higher doses of opioids and opioid-related mortality is apparent in both unadjusted and adjusted models.” However, using unpublished data, inaccessible to readers, as a defense is questionable and, at any rate, this neither supports a hypothesis that doses ≥200 mg/d MED are most hazardous nor that moderate doses are clearly distinct from 20-49 mg/d MED (which is not statistically different from <20 mg, as shown in the figure above).
- As is typical in research of this sort, outcomes are presented as odds ratios that are easily misleading and misunderstood by readers. For example, Gomes et al. claim that ≥200 mg/d MED doses were associated with nearly a 3-fold increase in the “risk of opioid-related mortality.” Actually, this was a 3-fold increase in the odds (OR=2.88) not the risk; the relative risk increase for ≥200 mg/d, which we calculated from data in the figure above, was 0.76 and the absolute risk difference between cases and controls in the study data for this category was only 0.10. [Interpreting risk in pain research data was discussed in an UPDATE here and problems with Odds Ratios were examined here.]
The authors further claim their study demonstrates “a substantial increase in the relative risk of opioid-related mortality associated with high opioid doses,[emphasis added]” but they do not calculate those risks for the reader. And, they acknowledge indirectly that, while absolute risks in the overall general population cannot be determined by their study methodology, such risks would likely be small. Which brings into question the rationale behind their argument for this study’s relevance or importance.
- When examining end points such as deaths attributable to a pharmacotherapy, randomized controlled trials are inappropriate and likely unethical. However, retrospective case-control studies such as this pose many challenges, particularly in the selection of adequate subjects to serve as controls. In his letter, Williamson observes, “It is apparent that the authors had difficulty matching cases and controls. The authors acknowledged the inability of the disease risk index to control for important confounders by stating ‘as expected, cases and controls differed on baseline characteristics associated with the risk of addiction and drug-related adverse events.’ There was also a significant decrease in matched controls with increasing dose categories” [Williamson 2011]. Therefore, Williamson contends, the ability of statistical techniques to adjust for imbalances in the many confounding factors between groups must be questioned and the true contribution of opioid dose as a cause of death is dubious.
In their response letter, Gomes et al. [2011B] acknowledge that “we did not report the proportion of cases in each dose group that were successfully matched in our original article, but in fact the proportion of cases fully matched to 4 controls did not differ substantially among groups (range, 81.1% to 84.1%, including 83.2% among people prescribed 200 mg/d of MED).” Besides referring to unreported data as a justification, again, the authors seem to be conceding that there was indeed a degree of mismatching across all groups. While sophisticated statistical manipulations can be, and were, used to adjust for this, as well as for an extensive list of potential confounders other than opioid dose that might have affected outcomes, at some point it must be considered that statistical maneuvering can only go so far before a study lacks validity.
In an invited commentary accompanying the article by Gomes et al., Mark Sullivan, MD, PhD, with the University of Washington School of Medicine [Sullivan 2011], notes that the evidence of mortality risks inherent in high-dose opioid therapy in the present study is consistent with earlier investigations, such as that by Dunn and colleagues. However, we have previously assessed fallacies of the Dunn et al. investigation [here], and warned that it might someday be wrongly touted as evidence to support similarly flawed research — as it is by Sullivan.
Both Gomes et al. and Sullivan note that mortality risks were observed to be elevated at doses exceeding thresholds in recent clinical guidelines. One of these that they reference is from the APS/AAPM (American Pain Society / American Academy of Pain Medicine), which actually does not advocate against higher opioid dosing and, in fact, acknowledges that there is insufficient evidence to support their own consideration of ≥200 mg/d MED as being a high dose — in short, the 200 mg/d threshold was just a “best guess” of the guidelines panel based on very limited evidence of any quality.
Then, using both the Gomes et al. and Dunn et al. studies as a justification, Sullivan notes how his own state, Washington, has adopted a 120 mg/d MED benchmark ceiling of opioid prescribing for chronic noncancer pain, and higher doses would require special considerations or actions by healthcare providers prior to continuing therapy [described more in an UPDATE here]. However, even while imposing this rule, state authorities waffled on this point by conceding that there is no predefined ‘safe’ opioid dose and that >120 mg/d MED is not necessarily unsafe.
What is clear is that the so-called evidence for increased mortality risks of opioids at arbitrarily determined higher doses is unclear, or at best weak. Yet, many researchers and even legislators and government agencies seem to accept that a collection of weak evidence amounts to strong proof that favors putting limits on opioid dosing. This faulty and unscientific reasoning brings us to some of the most disconcerting points expressed in these articles…
- Sullivan states in his commentary that, even though absolute risks of mortality associated with high-dose opioids is low, “death due to therapy for a nonprogressive, nonfatal condition must be taken seriously” [emphasis added]. Which implies that chronic pain is neither a progressive nor fatal condition that justifies the alleged risks of higher-dose opioid therapy affording pain relief.
To say the least, this is an uninformed, biased, and detrimental perspective. Prior research has demonstrated a profound link between severe chronic pain and death; inflicting nearly a 70% greater mortality risk than even cardiovascular disease [see UPDATE here]. And, in his letter to the editor, Williamson observes that evidence of an association between suicide rates and pain severity would suggest that chronic pain can indeed be fatal. Similarly, the association of unrelieved chronic pain and suicide also was discussed in UPDATES [here] and [here].
- However, in their rebuttal letter, Gomes et al. state: “[W]e agree with Dr. Williamson that chronic pain may be so severe that it increases the risk of suicide. However, given the absence of evidence that high-dose opioid therapy reduces the risk of suicide and the presence of evidence that many individuals commit suicide with opioids, the risk of suicide is yet another reason to prescribe opioids particularly cautiously to individuals with chronic nonmalignant pain” [emphasis added]. The implicit logic of this argument is abhorrent; that persons with chronic pain — possibly severe and unrelieved pain — may use their opioid medications to commit suicide and, therefore, they should be prescribed lower doses or denied opioids altogether!
- Lending a voice of greater reason, Williamson concludes his letter by stating, “It is important to ensure the safety and efficacy of any treatment for chronic pain by identifying and managing patient risk factors for adverse outcomes. However, it is equally important not to set arbitrarily limits on a given treatment if it can improve quality of life for patients with this common and poorly treated condition.”
Besides ramifications for negatively affecting the well-being of patients with chronic pain, another serious concern about this sort of research is that, with the increasing implementation of electronic medical records and the vast reservoirs of data that they represent, we will probably see ever more investigations taking advantage of data mining, often called “data dredging,” techniques. We have previously expressed concerns [here] that these approaches could produce a flood of misleading “pseudoscience” that might be more of a hindrance than a help in furthering better care for patients with pain.
Many questions about opioid safety — patient selection, maximum effective dosing, toxicity and associated mortality, etc. — are difficult to approach using high-quality research designs, such as randomized controlled trials. However, methodologies like data-mining generally provide only “soft answers” to those “hard questions,” and we should not accept these studies as valid or incontrovertible evidence. It is important that readers understand the limitations of these studies, can assess their flaws, and put the results into proper perspective to reach their own conclusions.
> Gomes T, Mamdani MM, Dhalla IA, et al. Opioid Dose and Drug-Related Mortality in Patients With Nonmalignant Pain. Arch Intern Med. 2011A;171(7):686-691 [abstract here].
> Gomes T, Mamdani MM, Dhalla IA, et al. Opioids and Dose-Related Deaths—Association or Causation?—Reply [Letter]. Arch Intern Med. 2011B;171(18):1688-1689 [extract here].
> Sullivan MD. Limiting the Potential Harms of High-Dose Opioid Therapy: Comment on "Opioid Dose and Drug-Related Mortality in Patients With Nonmalignant Pain." Arch Intern Med. 2011;171(7):691-693 [extract here].
> Williamson OD. Opioids and Dose-Related Deaths—Association or Causation? [Letter]. Arch Intern Med. 2011;171(18):1687-1688 [extract here].