In his letter, Cryer claims that our results are incorrect for reasons of confounding by indication. Undeniably, indication bias is present between the cohorts of acetaminophen users and NSAID users because, as we stated in the article, sicker patients were more commonly prescribed the safer drug (acetaminophen). Propensity scores (1) and multivariate models were used to adjust for this bias. The propensity score was the likelihood of being prescribed an acetaminophen versus an NSAID based on baseline risk factors. High propensity scores reflected high risk levels. Correcting for propensity scores means comparing patients with similar baseline risk factors. For example, patients in the high propensity score quintile (>80%) are all at high risk of adverse GI events, most likely because of prior GI events and use of anticoagulants. If we compare the rate of GI events in patients who received low doses versus high doses of acetaminophen within this quintile, we would be comparing patients at high risk of GI events who received a low dose of acetaminophen versus those who received a high dose of acetaminophen. Residual bias may remain, because models do not fully control for all confounders. However, the arguments that Cryer puts forward to justify his claim are not valid.
First, Cryer states that the acetaminophen cohort would have had a higher incidence of GI events even in the absence of exposure to acetaminophen. This is true, but it proves that the baseline risk level of the acetaminophen cohort is higher than that of the NSAID cohort, but it does not justify the difference in the incidence of dyspepsia between the low-dose and high-dose acetaminophen patients.
Second, Cryer states that it is likely that patients who received higher dosages of acetaminophen were also those who were older and had higher rates of a history of GI events and use of concomitant medications. We looked at the distribution of acetaminophen dosage in all propensity score quintiles. As we stated in the article, this distribution was similar. Therefore, high-dose acetaminophen was not preferentially prescribed in the high-risk group. The proportions of patients prescribed a high dose of acetaminophen were similar in all risk groups.
Third, according to Cryer, it is possible that the rate of dyspepsia in the NSAID cohort was artificially lowered by the higher rate of prescriptions for gastroprotective agents (GPA). The higher rate of GPA use in the NSAID cohort could have lowered the incidence of GI events in that cohort; however, in the model, GPA use (including that for prophylaxis) was counted among the GI events. The higher rate of GPA use for prophylaxis would offset the decrease in the GI event rate due to this use, therefore disadvantaging NSAIDs.
Fourth, Cryer claims that we did not describe whether patients receiving higher dosages of acetaminophen had higher rates of a history of ulcer disease. As we stated above, high-dose acetaminophen was given to patients independent of their prior level of risk for GI events (“A similar distribution of acetaminophen and NSAID dosage across propensity quintiles was observed”).
Fifth, Cryer states that followup complications of a GI bleed would recur within the first 2 weeks of the initial GI bleed, and, therefore, that the observed GI events would be recurrent events due to a prior bleed. Again, this would increase the rate of baseline events in the acetaminophen patients but would not explain the difference observed between patients receiving high-dose acetaminophen and those receiving low-dose acetaminophen. Moreover, in a sensitivity analysis, we excluded patients with prior GI events, cancer, NSAID use, or acetaminophen use, and the results between high-dose versus low-dose acetaminophen did not change.
Finally, our report is consistent with the literature. Studies that examined higher dosages of acetaminophen (4,000 mg/day) reported similar levels of subjective gastric discomfort symptoms in patients receiving ketoprofen and those receiving acetaminophen (2) and a similar incidence of adverse events in the ibuprofen and acetaminophen groups (3). Several randomized controlled trials and epidemiologic studies have shown an increase in the rate of dyspepsia in patients who received acetaminophen (4–6). Dyspepsia often is not caused by an erosion and is related to patient satisfaction and relief.
We thank Bannwarth for his interest in the study. In most of our previous work concerning GI events associated with NSAIDs, we used patients receiving acetaminophen as a control group, and we continue to do so. We do not think that this article contradicts our previous findings. It only indicates that the baseline costs observed in our previous study (7) were overestimated, because some of the GI-associated costs would have been attributable to high-dose acetaminophen and were counted as baseline costs. Therefore, the costs attributable to NSAIDs were probably higher than those reported. In our previous report, we considered all patients prescribed acetaminophen as one group and did not compare the costs for low-dose versus high-dose acetaminophen. Lower dosages of acetaminophen are more prevalent than were higher dosages. In the present study, ∼17,000 patients were receiving ≤2,600 mg/day and 4,000 were receiving >2,600 mg/day. Therefore, in our previous report, the results in the acetaminophen cohort would have been driven by the lower-dose patients. Moreover, in the previous study, we included only patients with no prior cancer, GI events, or NSAID or acetaminophen use. Therefore, the group would be comparable with that considered in the first sensitivity analysis of the current paper.
In comparing these 2 groups, we find that the results of the 2 studies were similar. In the current study, the risk ratio in patients with no previous use of NSAIDs or acetaminophen and no history of GI events or cancer who were taking high-dose acetaminophen was 0.8 when compared with NSAIDs users and 1.8 when compared with users of low-dose acetaminophen. Therefore, if we do the mathematics, the risk in the acetaminophen group will be half of that in the NSAID group. New users of acetaminophen are, perhaps, prescribed more prophylaxis than are patients who continuously use these agents, a phenomenon called depletion of susceptible individuals over time. Patients who used NSAIDs and did well are likely to continue doing well, at least in the short term.
In conclusion, we agree with both authors that acetaminophen is safer than NSAIDs, but perhaps patients who do not experience relief with regular doses of acetaminophen should be monitored more closely.