Results
Additional Follow-up and Events Identified
Follow-up for mortality was 25,833 person-years, including an additional 328 person-years identified during the reevaluation effort from participants with incomplete vital status at the end of the original RECORD study. Follow-up for the CV death, MI, or stroke composite was 23,692 person-years because participants without an event were censored on the date of the last face-to-face study visit. Vital status follow-up was complete for 96.0% of the participants (96.7% and 95.4% in the rosiglitazone and metformin/sulfonylurea groups, respectively). For the CV death, MI, or stroke composite end point, follow-up was complete for 83.3% of the participants (84.5% and 82.1% in the rosiglitazone and metformin/sulfonylurea groups, respectively).
DCRI CEC Metrics
Querying of the original RECORD database identified 314 deaths, 2,101 suspected MIs, and 496 suspected strokes. Of these, the computer program screening the database identified 2,052 suspected MIs and 468 suspected strokes, and manual review of documents by DCRI CEC coordinators identified an additional 49 suspected MIs and 28 suspected strokes. Of the 2,911 suspected events identified, 701 nonfatal events were not processed because they were duplicate events or no indication of an event was identified upon further processing. The remaining 314 deaths, 1,474 suspected MIs, and 422 suspected strokes were processed and adjudicated by the DCRI CEC clinicians.
Clinical Outcomes
Table I shows the CV death, MI, and stroke outcomes using the DCRI CEC results and the original RECORD CEC results by treatment using the original RECORD end point definitions. Figure 1 shows Kaplan-Meier curves for the primary composite outcome using DCRI CEC results and the original RECORD end point definitions. A total of 184 CV or unknown cause deaths (88 rosiglitazone, 96 metformin/sulfonylurea), 137 MIs in 128 participants (68 rosiglitazone, 60 metformin/sulfonylurea), and 119 strokes in 113 participants (50 rosiglitazone, 63 metformin/sulfonylurea) were identified by the DCRI CEC efforts using the same end point definitions as used in the original RECORD CEC. For the primary end point—time to first occurrence of CV (or unknown cause) death, MI, or stroke—no statistically significant difference was observed between rosiglitazone and metformin/sulfonylurea using the original RECORD end point definitions (HR 0.95, 95% CI 0.78–1.17). These results were comparable with those reported in the original RECORD study (HR 0.93, 95% CI 0.74–1.15). Cardiovascular (or unknown cause) death was similar between rosiglitazone and metformin/sulfonylurea using the original RECORD end point definitions (HR 0.90, 95% CI 0.68–1.21). For time to first fatal and nonfatal MI and for time to first fatal and nonfatal stroke, the comparisons between rosiglitazone and metformin/sulfonylurea were HR of 1.13 and 95% CI of 0.80 to 1.59 and HR of 0.79 and 95% CI of 0.54 to 1.14, respectively.
(Enlarge Image)
Figure 1.
Kaplan-Meier curves for the primary composite end point using the DCRI CEC results and the original RECORD end point definitions.
The HR for all-cause mortality was the same using the DCRI CEC and original RECORD CEC results (0.86, 95% CI 0.68–1.08).
Original End Point Definitions Compared With FDA Definitions
Table II shows composite CV death, MI, and stroke outcomes and the contribution of each event to the composite from the DCRI CEC results using the original RECORD end point definitions and the new FDA definitions. Figure 2 shows the Kaplan-Meier curves for the CV death, MI, and stroke composite end point using the DCRI CEC results and the new FDA definitions. The event rates and treatment comparisons are similar using the original RECORD end point definitions and the new FDA end point definitions. A comparison at the event level rather than the patient level (as in Table II) of the cause of death adjudications, number of MIs, and number of strokes using the original RECORD end point definitions and the new FDA end point definitions is shown in Table III. The results of cause of death adjudication were the same using both sets of definitions. The use of the new FDA end point definitions resulted in a small number of additional stroke (4) and MI (14) events primarily because the new FDA end point definitions for MI or stroke do not require hospitalization.
(Enlarge Image)
Figure 2.
Kaplan-Meier curves for the primary composite end point using the DCRI CEC results and the new FDA end point definitions.
Sensitivity Analyses
Table IV shows the results of a series of sensitivity analyses using the DCRI CEC results with the new FDA end point definitions. In the on-treatment analyses and landmark analyses attempting to account for potential bias after the interim publication, no evidence of heterogeneity in treatment effect was observed compared with the overall DCRI CEC results. Additional analyses using data acquired before amendment 7 or using an end date of the earliest possible final study visit showed similar results (data not shown).
Of the events adjudicated by the DCRI CEC group using the new FDA definitions, 45.5% of all deaths, 78.1% of MIs, and 94.3% of strokes were determined to have sufficient information by the CEC physicians. Results of the analyses of only those events with sufficient information were consistent with the analyses for the composite of CV death, MI, and stroke and the individual components using the DCRI CEC results.
Comparison of the Original RECORD CEC and the DCRI CEC Results
Figure 3 details the death reporting of the original RECORD CEC analysis and the DCRI CEC analysis. One participant was reported dead in the original data set, but the DCRI CEC group did not believe that there was enough evidence to confirm that the participant died. In total, the DCRI CEC group identified 20 deaths not reported in the original RECORD report. Of these 20, 8 (5 rosiglitazone and 3 metformin/sulfonylurea) occurred before the DCRI-defined final follow-up of December 31, 2008, and 12 (7 rosiglitazone and 5 metformin/sulfonylurea) occurred after this date. Of the 8 deaths occurring before December 31, 2008, 4 occurred before the end of the original RECORD-defined final follow-up. Seven of the deaths occurred during the additional 328 person-years of follow-up obtained during the reevaluation effort.
(Enlarge Image)
Figure 3.
Deaths reported by the DCRI CEC and original RECORD CEC and reasons for differences.
Overall, the DCRI CEC effort reported more CV or unknown deaths (184 vs 131), more MI events (128 vs 120), and more strokes (113 vs 109) compared with the original RECORD CEC process. Although numbers of all-cause deaths increased modestly with the DCRI CEC effort, the proportion of deaths classified as CV death by the DCRI CEC, which included deaths of unknown cause, was notably higher (see Table I). Most of this was explained by the DCRI CEC group adjudicating deaths with little information or deaths in participants with known cancer but no details about cancer progression available in the months before reported death as unknown cause because patients with cancer may die of many reasons besides cancer; these deaths were then included in the CV cause of death analyses, as specified in the DCRI and original RECORD analytic plans. Because the original RECORD CEC reported these as non-CV death, they were not included in the CV cause of death analyses.
Figure 4 shows the agreements and disagreements in the classification of cause of death for the 251 events adjudicated by both the original RECORD CEC and the DCRI CEC using the original RECORD cause of death end point definitions. Overall, there was agreement for 199 (79%) and disagreement for 52 (21%).
(Enlarge Image)
Figure 4.
Agreements and disagreements in the classification of cause of death between the original RECORD CEC and the DCRI CEC.
It was not possible to determine the total number of agreements or disagreements between the 2 CEC groups for MI or stroke because the original RECORD CEC reviewed only hospitalizations to identify MI and stroke events, whereas the DCRI CEC reviewed any suspected MI and stroke events. However, 14 MI events were reported by either the original RECORD CEC or the DCRI CEC but not both. Of these 14 MI events, 11 (5 rosiglitazone and 6 metformin/sulfonylurea) were reported by the DCRI CEC, and 3 (1 rosiglitazone and 2 metformin/sulfonylurea) by the original RECORD CEC. In total, 18 stroke events were reported by either the original RECORD CEC or the DCRI CEC but not both. Of these 18 stroke events, 11 (7 rosiglitazone and 4 metformin/sulfonylurea) were reported by the DCRI CEC, and 7 (3 rosiglitazone and 4 metformin/sulfonylurea) by the original RECORD CEC.