??tramadol for OA inc mortality/probs with statistic

A recent article suggested that tramadol use in patients with osteoarthritis was associated with increased mortality (see tramadol inc mortality jama2019 in dropbox, or doi:10.1001/jama.2019.1347). However, there are big questions about the validity of the statistical methodology used.

Details:

-- 252,859 patients from the Health Improvement Network database in the UK who were at least 50 years old and had a diagnosis of osteoarthritis were evaluated: 164,923 on tramadol and 87,936 on other medications

    -- these patients were seen between 2000-2016, and had not been on the OA medications below for one year before entering the study

-- original cohort: mean age of those on tramadol was 71, mean age of others was 68, socioeconomic deprivation index score was slightly higher in those on tramadol 2.6 vs 2.4, BMI on tramadol was 29 vs 28, duration of osteoarthritis was 7 years vs 6 years, visits to GP (general practitioners) 14 vs 11)

-- past alcohol use was higher on tramadol (3% vs 2%)/current use was lower (74% vs 79%), current smokers: no difference, CKD in tramadol group was skewed to more stage III to stage V, hypertension 55% vs 42%, ischemic heart disease 20% vs 12%, hyperlipidemia 18% vs 14%, anxiety 16% vs 14%, diabetes 16% vs 10%, angina 13% vs 8%, peptic ulcer 7-8% in both groups, atrial fibrillation 9% vs 3%, MI 7% vs 4%, heart failure 5% vs 3%

-- medication differences: tramadol group had significantly more on antihypertensives, statins, aspirin, other opioids (41% vs 26%)

-- after propensity score matching, 88,902 patients were included in the analysis, and were well-matched between the 2 groups, with mean age 70, 61% women:

    -- tramadol: 40451 people

    -- naproxen: 12,397 people

    -- diclofenac: 6512 people

    -- celecoxib: 5674 people

    -- etoricoxib: 2946 people

    -- codeine: 16,922 people

Results:

-- during one year follow-up,

    -- tramadol: 278 deaths (23.5/1000 person-years)

    -- naproxen: 164 deaths (13.8/1000 person-years), 71% increase vs tramadol with HR 1.71 (1.41- 2.07)

    -- diclofenac: 121 deaths (19.2/ 1000 person-years), 88% increase vs tramadol with HR 1.88 (1.51-2.35)

    -- celecoxib, 102 deaths (12.8/1000 person-years), 70% increase vs tramadol with HR 1.70 (1.33-2.17)

    -- etoricoxib, 37 deaths, twice the mortality vs tramadol with HR 2.04 (1.37-3.03)

    -- codeine, 552 deaths, no significant difference with HR 0.94 (0.83-1.05)

-- causes of death (though 16 to 30% of causes of death could not be ascertained):

    -- tramadol vs naproxen: twice as many infections or cancer-related deaths

    -- tramadol vs diclofenac: more than twice as many cancer or respiratory deaths

    -- tramadol vs celecoxib: more than twice as many infections, cancer, or respiratory deaths

    -- tramadol vs etoricoxib: between 4 and 7 times as many cardiovascular or respiratory deaths

Commentary:

-- in this trial, those on tramadol were older, had higher BMI, longer duration of osteoarthritis, higher prevalence of comorbidities (chronic kidney disease, diabetes, hypertension, cardiovascular diseases) and other meds (other NSAIDs, other opioids, aspirin, statins, antihypertensives, and antidiabetics) and more healthcare utilization (visits to GP)

-- there are not many safe and effective treatments for patients with osteoarthritis. NSAIDs are associated with a wide range of potentially serious adverse effects, including renal failure, heart failure, hypertension, and gastritis/GI bleeds; and these adverse effects are of particular concern in older people, exactly the ones who have osteoarthritis. And acetaminophen only does so much....

-- the most recent 2013 guidelines from the American Academy of Orthopedic Surgeons does strongly recommend tramadol or NSAIDs for symptomatic osteoarthritis

-- the most recent 2012 guidelines from the American College of Rheumatology also recommends tramadol as first-line therapy for knee osteoarthritis, along with acetaminophen, NSAIDs, and corticosteroid injections

-- likely as a result of these types of recommendations and concerns of NSAID adverse effects, as well as the fact that tramadol is a weak opioid, there has been a significant increase in tramadol prescriptions in the US

------------------------------------

a subsequent editorial strongly challenged these results (see statistics propensity matching jama2020 in dropbox, or doi:10.1001/jama.2019.21558)

--the above article used the most common variant of propensity match scoring, “nearest-neighbor”, where they scroll through the patients to find the closest matches and throw out those without matches

--in the above case, comparing tramadol vs diclofenac, for example:

    --prior to propensity score matching, tramadol vs diclofenac:

        --mean age 72.1 vs 67.5

        --number of visits to GP: 14.3 vs 9.7

    --after propensity matching, only 38% of eligible patients were able to be matched, as 6512 pairs:

        --mean age 70.3 vs 72.1

        --number of visits to GP: 12.3 vs 14.3

    --ie, after propensity matching: instead of patients being older on tramadol by 4.6 years and seeing the GP 4.6 more visits, propensity matching led to the tramadol patients being younger by 1.8 years and seeing their GP 1.5 fewer visits!!!!

Commentary:

-- propensity match scoring is a basic methodology to try to equalize mathematically disparate groups of patients in observational studies, and it is important in our being able to compare outcomes in different observational studies more incisively

-- BUT, there are some concerns in the original tramadol trial would make their results open to questioning:

    -- as in the above editorial, the elimination of about 2/3 of patients in the course of propensity scoring is very concerning. It does suggest that there may be confounding by indication (eg those on tramadol were put on it because they were older and sicker, which explains why it was so hard to match the patients); this is a far cry from a random selection of patients, since the researchers are picking and choosing a small minority to evaluate, opening up real potential biases in the composition of the groups and their conclusions

    -- In addition, the lack of a specific association with tramadol and mortality: for some of the medications tramadol had significantly increased cardiovascular mortality, for some there was increased mortality from infections, for several it was an increase in cancer mortality, for several there was also increased risks of respiratory mortality. This makes it difficult to attribute a specific reason for tramadol-induced increased mortality (and the lack of a coherent plausible explanation makes it more difficult for us to accept the associations…), and also does suggest there might be residual fundamental differences between the groups after propensity scoring

        -- for example, was tramadol used more in those with some baseline CKD, or heart failure, or hypertension, or other unmeasured comorbidity/psychosocial issue, or simply in an older age group because of the large number of potentially serious adverse effects of NSAIDs?? And these tramadol users were fundamentally frailer individuals (and therefore had a higher mortality anyway)

        -- and, looking at the number of GP visits may be a better surrogate for sickness than just controlling for the presence of their comorbidities: visits to clinician probably correlates better with the intensity of illness vs just having that illness on their problem list. And in the above study, the tramadol group had more GP visits, but after propensity scoring had fewer!!!

    -- also, the overall use of tramadol and codeine was much more than NSAIDs, despite the more usual approach (as suggested in UpToDate, for example) of trying NSAIDs first, also suggesting this group may have had more severe OA and may be very different than those with milder cases. (in fact, NSAIDs were prescribed in 86% in the tramadol group, also suggesting that these patients had more advanced and more difficult to treat osteoarthritis requiring more than just tramadol.  Perhaps those on tramadol may not have been able to exercise much. And that this lack of exercise itself is associated with more cancer and mortality (see http://gmodestmedblogs.blogspot.com/2019/03/5-10-min-of-exercise-may-improve-health.html )

    -- and, there was a significant difference in tramadol mortality within just 2 months, making it somewhat difficult to attribute the cancer mortality, for example, to tramadol (though we do not know specific breakdowns of these outcomes, and many of the outcomes were too infrequent to have statistical significance)

   --so, it is really hard to accept the conclusions of this study!!!

-- one of the major reasons I do so many steroid injections for osteoarthritis in patients (after trying acetaminophen and physical therapy) is that the long-term oral/systemic meds available have (it seems to me) much more likelihood of serious adverse effects than the potential ones from intra-articular lidocaine and steroids.  For my older patients who decline or do not respond well to injections, I do tend to use tramadol a lot (which also has benefit for those with neuropathic pain, if that also happens to be present) or codeine. see http://gmodestmedblogs.blogspot.com/2019/10/intra-articular-steroids.html

So, I personally do not ascribe any real value to the study showing increased mortality in those on tramadol, and that, from my perspective, following their distorted one-liner that tramadol is associated with increased mortality may deprive patients of effective systemic therapy and put many at risk for long-term major adverse events. though i still like the advantages of intra-articular steroids

And, one of the reasons I bring up this article is that it gives an opportunity to see how statistics can sometimes be used to reinforce wrong conclusions. Sort of like using the “p-value” as a marker of clinical significance:

    -- a p-value that is highly significant in a very large cohort may have little clinical utility: it may be really statistically significant, with p<0.0001, that the systolic blood pressure was 0.5mmHg lower by an intervention in a large sample. But is that clinically important??

    -- and a p-value of 0.051 in a study may actually reflect an important clinical finding: if there is a 40% decreased risk of a disease by an intervention in a smallish study, but the confidence intervals are from 0.25-1.01, this would not be "statistically significant" but way more likely than not is clinically significant. But, a very similar confidence interval of 0.25-0.99 would also clear the p-value threshold of statistical signifcance with perhaps just a few additional patients assessed.   [this is the reason i put confidence intervals in my blogs since that really gives a lot of information about the likely importance of the intervention]

    -- and, by the way, even the Am Statistical Assn is “moving to a world beyond p <0.05”

geoff​

If you would like to be on the regular email list for upcoming blogs, please contact me at gmodest@uphams.org

to get access to all of the blogs:

1. go to http://gmodestmedblogs.blogspot.com/ to see them in reverse chronological order

2. click on 3 parallel lines top left, if you want to see blogs by category, then click on "labels" and choose a category​

3. or you can just click on the magnifying glass on top right, then  type in a name in the search box and get all the blogs with that name in them

please feel free to circulate this to others. also, if you send me their emails, i can add them to the list