RSS Feeds
Questioning the Accuracy of a Recent Review of Osteoporosis Medications
IN RESPONSE:
The concerns of Ms. Kuhl and colleagues about our article and the report from which it was derived (1) can be grouped into 3 categories: questions about how meta-analysis could find a statistically significant association between a drug and an outcome when none of the individual studies reported a statistically significant association, questions about what outcomes and studies should have been included in the review, and errors of fact.
First, the fact that a meta-analytic pooled result shows a statistically significant association when none of the individual studies themselves did so should not be surprising: Pooling studies to increase the statistical power to find effects is one of the primary rationales for performing meta-analysis. For the risk for upper gastrointestinal perforations, ulcers, and bleeding events associated with ibandronate, we combined gastrointestinal hemorrhage and hemorrhagic gastritis into upper gastrointestinal perforations, ulcers, or bleeding for 1 of the 2 studies in question (2) and combined duodenal ulcer and stomach ulcer into this category for the other study (3). Also, we combined all doses of ibandronate into 1 ibandronate group. The pooled result was a statistically significant association between use of ibandronate and lower risk for upper gastrointestinal perforations, ulcers, and bleeding.
Regarding the association between teriparatide and cancer risk, Ms. Kuhl and colleagues are incorrect in their assertion that 2 of the 3 studies cited have an identical incidence rate for treatment and placebo. The incidence rates are the same for treatment and placebo in only 1 study (4). When the 20-mg and 40-mg teriparatide groups in Orwoll and colleagues' study (5) are combined, cancer occurred in 3 of 290 (1%) teriparatide recipients versus 3 of 147 (2%) placebo recipients. As Ms. Kuhl and colleagues noted, in the third study, risk for cancer among teriparatide recipients is significant compared with that in placebo recipients (6). Combining these studies resulted in a statistically significant risk for cancer among participants treated with teriparatide relative to those treated with placebo (relative risk, 0.49 [CI, 0.27 to 0.9]).
Second, an external group of technical experts set the scope of our review, which was limited to evaluation of the benefits of agents used to treat or prevent osteoporosis on 1 outcome: risk for fracture. We were also directed to focus on a select group of adverse events, specifically cardiovascular events, including myocardial infarction and stroke; thromboembolic events, including pulmonary embolism and venous thromboembolic events; malignant conditions, including breast, colon, and lung cancer and osteosarcoma; upper gastrointestinal events, including perforations, ulcers, bleeding, and esophageal ulcerations; osteonecrosis; inflammatory eye reactions, including uveitis and scleritis; and acute phase reactions. However, all adverse events reported in any study included in the report, including hot flashes, are detailed in Appendix C of the technical report (1).
Specifically regarding the inclusion of data from various studies pertaining to the relationship between raloxifene and breast cancer, we did include data from the MORE trial. We used data from the study by Grady and colleagues (7) rather than Ettinger and colleagues' publication (8) because the former was more recent (2004 vs. 1999). Also, whereas the Ettinger and colleagues' publication has more comprehensive data on breast cancer, it only reports a risk ratio and the overall number of women with breast cancer (not by tamoxifen vs. placebo group), which cannot be used in a pooled analysis. This study reported that breast cancer was less frequent among women receiving raloxifene than those receiving placebo (relative risk, 0.3 [CI, 0.2 to 0.6]).
The RUTH trial (9) did not meet our criteria for inclusion in our review because it did not report fractures as an outcome. This is a stated limitation of our methods. The hazard ratio of any breast cancer for raloxifene relative to placebo in this study was 0.67 (CI, 0.47 to 0.89).
The STAR trial (10) compared the effects of raloxifene and tamoxifen. This trial reported fracture outcomes, and we included it in our analyses. However, we inadvertently neglected to include the STAR data on breast cancer in our report. In the STAR study, breast cancer occurred in 248 of the 9875 participants treated with raloxifene and in 220 of the 9872 participants treated with tamoxifen. The risk for breast cancer for raloxifene relative to tamoxifen, on the basis of data from this study, was 1.13 (CI, 0.94 to 1.36). This will be added to the technical report.
With regard to Ms. Kuhl and colleagues' final concern, among the 500 studies included in our review, the authors did identify 1 error of fact, for which we thank them. In the study by Reid and colleagues (11), cited in the report in Appendix C, page C-70, we inadvertently inverted the data for the incidence of arthralgia for the placebo and risedronate groups. The correct numbers are 16% for placebo and 24% for the 5-mg risedronate group. The original pooled analysis for the risk for musculoskeletal events for risedronate relative to placebo, reported in Appendix E, page E-8, produced an estimate of 0.4 (CI, 0.29 to 0.54) and the corrected pooled analysis is 0.64 (CI, 0.47 to 0.87). This change, although insignificant in terms of our conclusions, will be made to the Agency for Healthcare Research and Quality report.
Ms. Kuhl and colleagues state that our report erroneously concludes that endometrial cancer is not an adverse effect for tamoxifen. Pooled analyses of the risk for endometrial cancer were not reported in either the article or the technical report because our technical advisory panel did not recommend this type of cancer as a priority for this report. Rather, we pooled all types of gynecologic cancer. The pooled risk for any gynecologic cancer for tamoxifen relative to placebo was 1.52 (CI, 0.97 to 2.41). This is reported in Appendix E, page E-23, of the technical report. The incidence of endometrial cancer in the study by Fisher and colleagues (12) for the tamoxifen and placebo groups are correctly reported in Appendix C, page C-125.
Article and Author Information
-
Potential Financial Conflicts of Interest: None disclosed.
