Article: Phenylpropanolamine and the risk of hemorrhagic stroke
Authors: Kernan WN, Viscoli CM, Brass LM, Broderick JP, Brott TB, Feldmann E, Morgenstern LB, Wilterdink JL, Horwitz RI
Publication: New Engl J Med. 2000;343(25):1826-1832.
Background: Phenylpropanolamine (PPA) is a synthetic sympathomimetic which until 2000 has been commonly found in over-the-counter (OTC) appetite suppressants and cold medications. Since 1979, more than than 30 case reports suggesting that PPA is associated with intracranial hemorrhage have been published.
Step 1: Assess the study design
Step 1a: What was the study hypothesis? Did it make sense?
A hypothesis is never formally stated, although given the background of PPA and the discussion of 3 specific aims, it could be assumed that the null hypothesis would be:
1) Among women, there is no association between hemorrhagic stroke and the use of appetite suppressants containing PPA and any first use of products containg PPA
2) Among men and women, there is no association between the use of PPA (as either an appetitie suppressant or cold remedy) and hemorrhagic stroke.
3) Among men and women, there is no association between type of exposure to PPA and hemorrhagic stroke.
Studying the relationship between PPA and hemorrhage stroke does make sense, given both the published case reports and the 22 reports of association to the FDA. No prior investigation on the topic with controls had been done.
Step 1b: What was the study design?
The investigators chose a case-control study design. The outcome of interest (hemorrhagic stroke) is rare which makes the study design necessary. A RCT or cohort study would not have been logistically or ethically feasible.
Step 1c: Were there comparison groups?
Yes, there were cases (n=702) and controls (n=1376). The authors were close to achieving their goal of 2 controls/case (96% cases had 2 matched controls). They also achieved the number of patients needed to have a power of 80% to detect a statistically significant difference between female cases and controls.
Step 1d: How similar were the comparison groups?
The investigators did a thorough job collecting and reporting data on different patient characteristics. There were significant imbalances between the cases and controls because cases were more often black and less educated. They were more likely to have a history of hypertension, a family history of hemorrhagic stroke, regular alcohol use, recent use of cocaine, and a lower body mass index. They were also less frequently exposed to NSAIDs and more frequently exposed to agents containg caffeine and/or nicotine.
Step 1e: Was the exposure clearly defined and capable of being reliably evaluated?
The authors do no not define the amount of PPA necessary to be considered an exposure. In terms of the time frame from exposure to outcome, the authors define a "window of exposure." Although explained in somewhat of a unclear manner, the "window of exposure" essentially means exposure to PPA within 4 days of the onset of symptoms which lead to being seen by a clinician and the diagnosis of hemorrhagic stroke. The lack of an exposure was clearly defined as no use of PPA within the prior 2 weeks. Exposure was most likely determined as reliably as possible through live or phone interviews by trained researchers. Verification of exposure occurred if the subject produced the actual PPA-containing medication or chose one from a book of photographs. Subjects were excluded if they could not communicate well enough for an interview within 30 days of the stroke.
Step 1f: Was the outcome clearly defined and capable of being reliably evaluated? Was it a clinically important outcome?
The outcome was clearly defined as subarachnoid hemorrhage seen on CT scan, xanthochromia on LP, or parenchymal hemorrhage seen on CT or MRI scan consistent with clinical symptoms. These findings are reliably evaluated, especially since the authors excluded any subjects with brain lesions or history of stroke. Hemorrhagic stroke is a clinically important outcome due to its associated significant morbidity and mortality.
Step 1g: Were the exposures and outcomes measured in the same way in the groups being compared?
Exposures were measured in the same manner in both cases and controls. The outcome of interest, hemorrhagic stroke, occurred in only cases, so it was not measured in controls. Case-control studies in general are susceptible to recall bias and investigator bias. The authors attempted to prevent recall bias by using a highly structured interview in which the subjects were blinded to the PPA hypothesis and interviewed within 30 days of the hemorrhagic stroke. (Although it is not clear that interviewing a shorter time after the outcome would decrease the tendency for recall bias.) Several strategies were used to reduce the possibility of selection bias: active case surveillance and objective determination of eligibility. 39 patients declined to participate (in 3 cases their primary MD refused their participation), so for the most part the patients selected were enrolled.
Step 1h: How good was follow-up? Was it appropiately long? Was there excessive loss to follow-up?
These questions only apply to RCTs and cohort studies.
Step 2: Assess the results
Step2a: How strong was the association between exposure and outcome?
The strength of inference will be less for this study because it had a case-control design and therefore there may have been confounding variables that were not recognized. The adjusted odds ratio (OR) was very high (ie strong association) in one particular group the investigators studied: 1.6% of women who had hemorrhagic strokes had used PPA-containing appetite suppressants compared with only 0.1% of the female controls. The OR was 16.58 (1.51-182.21) with p=0.02. The strength of association also appears high and significant in women who had any use of products containing PPA (OR 1.98 (1.00-3.90) with p=0.05). However, there was no association between the use of PPA-containing cold remedies and hemorrhagic stroke in women, so one reasons that the association between all PPA-containing products and hemorrhagic strokes in women resulted from the strength of association between the appetite suppressants and hemorrhagic stroke. The other situations examined ( both sexes + PPA-containing products--> hemorrhagic stroke, both sexes+1st time use of PPA-containing products-->hemorrhagic stroke, both sexes + PPA containing appetitie suppressants--> hemorrhagic stroke, both sexes+PPA containing cold remedies-->stroke, etc for men and women separately) either were not statistically significant or do not have strength of association (or both). It would be impossible to show a strength of association between men using PPA-containing appetite suppressants and hemorrhagic stroke because none of the men in the study took them. Despite the lack of statistical significance and/or strength of association among these other situations studied, the authors were impressed by some of the trends (eg they state that first time use of PPA-containing products in women may have an association with hemorrhagic stroke).
Step 2b: Was there a temporal relationship between the exposure and outcome?
The authors tried to establish this as well as they could by including as case patients only those who had taken PPA within 4 days of the hemorrhagic stroke. However, they do not mention what the duration of PPA use was in patients. Someone taking PPA daily for years who has a hemorrhagic stroke is quite different from someone who took it only the 2 days prior to their hemmorrhagic stroke. The association between PPA exposure and hemorrhagic stroke would seem much greater in the second situation. Although the authors do not report or control for the duration of PPA use, this was true across both cases and controls which would tend to at balance this confounding effect. They did perform a separate analysis of only patients who had been exposed to PPA within 24 hours of their hemorrhagic stroke and found that adjusted ORs were slightly higher but still not statistically significant. The authors also mention the potential for temporal-precedence bias which "...refers to a systematic error in which an exposure is counted although the exposure occurs after the onset of the disease under study, often in response to disease symptoms." One measure that the authors took to check for temporal-precedence bias was excluding all patients who had experienced sentinel headache from the analysis. There was no major distortion of the results in doing so. Sentinel headache was defined as headache prior to symptoms which lead to being seen by a clinician and being diagnosed with a hemorrhagic stroke.
Step 2c: Was there a dose-response gradient?
The results of this study support a dose-response gradient. Appetite suppressants contain higher amounts of PPA than cold remedies. The investigators found the largest association of hemorrhagic stroke with appetite suppressants. Also, appetite suppressants are thought to have a higher potential for abuse than cold remedies and therefore would most likely be taken in greater amounts. The authors also performed an analysis which showed that the OR relating hemorrhagic stroke and PPA exposure was higher in those taking doses above 75 mg, although these results were not statistically significant.
Step 2d: How precise was the estimate of risk?
The main "positive" result of the study was an OR of 16.58 for women using PPA-containing appetite suppressants acquiring hemorrhagic stokes. The 95% CI for this OR was 1.51-182.21, which is large (ie unprecise). However the lower end of the CI, 1.51, is still above 1, so we can be convinced that there was a positive relationship. None of the other relationships mentioned in step 2a was statistically significant and none of the estimates was precise. Men who used PPA containing products actually appeared to have a lower risk of hemorrhagic stroke (OR 0.62), which does not really make sense. The upper end of the CI (1.92) would suggest that we cannot exclude that men exposed to PPA were at higher risk for hemorrhagic stroke. The bottom line is that a larger number of subjects might have produced more precise estimates of risk or no risk for many of the subgroups studied.
Step 2e: Did the authors adjust for differences between comparison groups?
Yes and no. It was stated that "...only the level of education changed the adjusted odds ratio for the association with hemorrhagic stroke by more than 10 percent." The results were reported as adjusted for race, smoking status, presence or absence of hypertension, and level of education. The authors did not adjust for a family history of hemorrhagic stroke, regular alcohol use, recent use of cocaine, a lower body mass index, and or recent exposure to NSAIDs or caffeine or nicotine containing agents, because these entities did not change the adjusted OR by more than 10%. Together or separately, however, these could be confounding variables.
Step 2f: Was the statistical analysis appropriate?
Yes. 1st the appropriate sample size required to show an association between PPA exposure and hemorrhagic stroke with 80% power was calculated. Chi square and Fisher's exact tests were used to compare patient characteristics between groups in the 1st phase. Conditional logistic models for matched sets were used in the 2nd phase to estimate ORs with CIs. They then calculated unadjusted and adjusted ORs.
Step 3: Do you agree with the authors' conclusions?
The conclusion that "Among women between the ages of 18 and 49 years, the use of a product containing [PPA] as an appetite suppressant was associated with an increased risk of hemorrhagic stroke" is believable based on the study results. The authors mention numerous times that 1st use of PPA, PPA exposure from cold remedies, and PPA use in men as well are likely to be associated with hemorrhagic stroke, although their data neither supports nor negates these conclusions.
Step 4: How applicable are the results?
Step 4a: Did the risk apply to a narrow or broad population?
The risk applied to women between the ages of 18-49 taking PPA-containing appetite suppressants. The risk of hemorrhagic stroke may apply to a broader group of people, but this study did not demonstrate that.
Step 4b: What was the magnitude of the risk?
The magnitude of risk was estimated by the authors to be that 1 woman may have a hemorrhagic stroke due to PPA within a 3 day window for every 107,000 to 3,268,000 women who use PPA containing appetite suppressants. They used a daily incidence of hemorrhagic stroke of 0.6 per million for persons 35-54 years of age. This estimate was not particular for women and may in fact be lower in women. Still this is only an estimate based on their OR and CI. This would seem clinically important given the estimate that millions of Americans take PPA-containing products each month. Based on these estimates, at least 1-30 women would have a PPA-related hemorrhagic stroke a month or up to 400 PPA-related hemorrhagic strokes a year (1).
Step 4c: Should clinical practice be altered because of the results of this study?
There was a high strength of association (OR >16) with a study design susceptible to unknown confounders and biases. Eliminating ~400 potential PPA-related strokes a year would seem to be a mandate. Recommending patients not take PPA-containing appetite suppressants is not enough, because the medications are OTC and have some abuse potential. Therefore, it seems reasonable that the FDA would recommend removing PPA-containing appetite suppressants from the market. What may seem unreasonable is that the FDA announced its intention to withdraw all PPA-containing products from the market. PPA-containing cold remedies are widely used. The main alternative active ingredient, pseudoephedrine, as one author put it, "...is not itself without potential problems, including its use in the synthesis of illicit methamphetamine and reports of ischemic colitis."(2) Although PPA is the same molecular entity present in both cold remedies and appetite suppressants, the doses, user population and abuse potential differs between these 2 OTC medications.
References
1) Fleming GA. The FDA, regulation and the risk of stroke. New Engl J Med. 2000;343:1886-1887.
2) Ernst ME, Hartz A. Phenylpropanolamine and hemorrhagic stroke. New Engl J Med. 2001;344(14):1094.