Formulary evidence sheet · historical exercise

Rofecoxib versus naproxen in VIGOR: stomach benefit and heart comparison

Information as of February 8, 2001. Uses only the article abstract (VX1) and the FDA cardiovascular review (VX2).

Policy call

Disclose a cardiovascular warning on the label: Yes.

As of February 8, 2001: yes. Labeling that presents the VIGOR gastrointestinal results should also disclose the VIGOR cardiovascular results, in the study description and in the Warnings section. This matches the FDA reviewer's recommendation (memo p. 37 (PDF p. 39), VX2).

The gastrointestinal benefit does not warrant ignoring the myocardial infarction comparison, and the naproxen-protection explanation cannot explain it away with these data. Committee materials that cite the stomach benefit should show the heart comparison beside it, each in its own unit.

Naproxen-protection explanation: plausible, not established, and not distinguishable in this trial from a rofecoxib-related increase. Absolute effect of either drug versus no therapy: unknown.

Historical evidence exercise. Not medical advice for any patient.

Heart versus stomach, as each source reports it

Stomach: confirmed GI events (VX1)

Rofecoxib: 2.1 per 100 patient-years
Naproxen: 4.5 per 100 patient-years
Article's relative risk, rofecoxib to naproxen: 0.5 (95 percent CI 0.3 to 0.6; P<0.001)
Direction: Fewer GI events on rofecoxib.

Heart: adjudicated acute MI (VX2 counts)

Rofecoxib: 20 of 4,047 randomized: 4.94 per 1,000
Naproxen: 4 of 4,029 randomized: 0.99 per 1,000
Crude risk ratio, rofecoxib to naproxen: 4.98 (approximate 95 percent CI 1.70 to 14.6, computed here)
Excess on rofecoxib: 3.95 per 1,000 randomized participants in this trial
Direction: More MI on rofecoxib.

Units differ. The GI figures are event rates per 100 patient-years of follow-up. The MI figures are crude risks: participants with an MI per randomized participant over the trial (median follow-up 9.0 months, VX1). The inputs give no GI event counts or patient-year totals, so this sheet does not convert one unit into the other or subtract heart from stomach.

The article's own abstract also reports the heart signal: MI in 0.4 percent of rofecoxib patients versus 0.1 percent of naproxen patients, relative risk 0.2 stated with naproxen on top (VX1). With rofecoxib on top, as for the GI result, that is about 5.

MI events per 1,000 randomized, by arm

Crude acute myocardial infarction per 1,000 randomized participants. Zero baseline. Counts from the FDA reviewer table, memo p. 18 (PDF p. 20) (VX2). Excess on rofecoxib: 3.95 per 1,000 randomized in this trial.

Missing third arm: no therapy or placebo. Not in this trial. MI risk without either drug is unknown, so the chart cannot show whether rofecoxib raised risk, naproxen lowered it, or both. Only the gap between the two bars is measured.

Chart shows the two trial arms only. Open the reviewer reasoning below to add what the trial cannot measure.Chart now marks the missing no-therapy comparison, which this trial cannot supply.

OpenClose FDA reviewer reasoning: could naproxen protection explain the gap?

“The VIGOR study was a large study with a longer drug exposure and follow-up than the two smaller studies (085 and 090). The cardiovascular thrombotic event rates, while not high, were significantly different between the two groups; most striking were the myocardial infarction event rates.”
FDA reviewer memo, memo p. 34 (PDF p. 36), VX2

“As there is no placebo group, it will be difficult to assess the CV thrombotic risk with rofecoxib use compared with no therapy at all.”
FDA reviewer memo, memo p. 34 (PDF p. 36), VX2

“It would be difficult to imagine inclusion of VIGOR results in the rofecoxib labeling without mentioning cardiovascular safety results in the study description as well as the Warnings sections.”
FDA reviewer memo, memo p. 37 (PDF p. 39), VX2

Reading for the committee. In these excerpts the reviewer does not treat any explanation as settled. The reviewer notes there is no placebo group, so risk versus no therapy cannot be assessed, and still recommends that labeling mention cardiovascular results in the study description and Warnings. A naproxen-protection explanation is therefore not established. Plausible only as a hypothesis, and impossible to distinguish here from a rofecoxib-related increase, or from a mix of both.

Naproxen protection: established, plausible, or indistinguishable?

Not established. Plausible only as a hypothesis, and impossible to distinguish here from a rofecoxib-related increase, or from a mix of both.

With two active arms and no placebo arm, the trial measures only the difference between rofecoxib and naproxen. A lower naproxen risk and a higher rofecoxib risk produce the same comparison. The inputs contain no data that test naproxen protection directly (VX1, VX2).

Whichever explanation holds, the measured comparison stands: in this randomized trial, participants assigned rofecoxib had more adjudicated MIs than participants assigned naproxen. One trial with only 4 events in the naproxen arm cannot rule out chance entirely. This sheet does not claim proof of causation.

Audit of the FDA table's other cardiovascular rows

Source note, quoted: “Patients may be counted in more than 1 row, but are only counted once within a row.” (memo p. 18 (PDF p. 20), VX2). Rows overlap, so each row is read on its own and none are added together.

Each row on its own; rows are not additive. Denominators: rofecoxib 4,047, naproxen 4,029 randomized.
FDA row	Rofecoxib	Naproxen	Crude ratio, rofecoxib to naproxen	Difference per 1,000 randomized
Acute Myocardial Infarction	20 of 4,047; 4.94 per 1,000	4 of 4,029; 0.99 per 1,000	4.98	+3.95
Any Event	47 of 4,047; 11.61 per 1,000	20 of 4,029; 4.96 per 1,000	2.34	+6.65
Cardiac Events (Fatal/Nonfatal)	28 of 4,047; 6.92 per 1,000	10 of 4,029; 2.48 per 1,000	2.79	+4.44
Cardiovascular Death	6 of 4,047; 1.48 per 1,000	6 of 4,029; 1.49 per 1,000	1.00	-0.01

Nesting check within each arm (acute MI, then cardiac events, then any event): rofecoxib: 20 ≤ 28 ≤ 47 (consistent); naproxen: 4 ≤ 10 ≤ 20 (consistent). Cardiovascular death is 6 vs 6, in line with the article's statement that cardiovascular death was similar; the MI gap is not mirrored in cardiovascular deaths. Any event (47 vs 20) and cardiac events (28 vs 10) point the same way as MI, but they can include the MI participants, so they are not independent confirmation.

Article versus FDA counts

Reconciliation of the article abstract with the FDA reviewer table
Item	Article, VX1	FDA counts, VX2	Finding
Randomized denominators	8,076 randomized	4,047 rofecoxib + 4,029 naproxen = 8,076	Match. The FDA arm sizes sum to the article's randomized total.
Rofecoxib MI percentage	0.4 percent	20 of 4,047 = 0.49 percent	Does not match at one decimal place: the FDA count gives 0.5 percent. The FDA table is labeled updated application data; these inputs do not show which events account for the difference.
Naproxen MI percentage	0.1 percent	4 of 4,029 = 0.10 percent	Consistent at one decimal place.
Direction of the reported MI relative risk	0.2 with naproxen as numerator (95 percent CI 0.1 to 0.7)	rofecoxib over naproxen, crude: 4.98	The article states the GI relative risk with rofecoxib on top (0.5) but the MI relative risk with naproxen on top (0.2). Read with rofecoxib on top for both outcomes, the article's MI figure is about 5 (CI about 1.4 to 10), the same direction and similar size as the FDA counts.
Cardiovascular death	rate of death from cardiovascular causes similar	6 vs 6	Consistent: equal counts in arms of nearly equal size.

Design facts: 8,076 randomized; two active arms; placebo arm absent (VX1, VX2).

What remains unknown

Causal counterfactual: unknown. What would MI risk have been in these patients on no therapy (or placebo)? The absolute effect of either drug on MI cannot be identified from this two-arm comparison. The 3.95 per 1,000 excess is a between-arm difference in this trial, not an attributable count of harmed people.

MI risk on no therapy or placebo in this population (causal counterfactual).
GI event counts and patient-year totals: not in the inputs, so GI rates stay per 100 patient-years.
Which MI events separate the article's 0.4 percent from the FDA's 20 of 4,047.
How results carry to other populations, doses or durations: only VIGOR's RA population and doses were studied.
Precision: only 4 MI events in the naproxen arm, so the ratio is unstable.

Not claimed on this sheet

Proof of causation: this two-arm study shows a between-arm difference, not proof that either drug changed MI risk.
A safe-alternative label for either drug.
A count of harmed people for any wider population.
Any placebo or no-therapy comparison.
Any event after February 8, 2001.

Source ledger

VX1: Comparison of Upper Gastrointestinal Toxicity of Rofecoxib and Naproxen in Patients with Rheumatoid Arthritis
Publisher
New England Journal of Medicine
Dated
2000-11-23
Used here
Abstract: randomized total, doses, median follow-up, GI rates per 100 patient-years, article MI percentages and relative risk.
Links
PubMed abstract record Publisher page via DOI
VX2: Review of cardiovascular safety database; NDA 21-042 S-007
Publisher
FDA Division of Cardio-Renal Drug Products
Dated
2001-02-01
Used here
Reviewer memo dated February 1, 2001: adjudicated event counts and arm sizes (memo p. 18), assessment (memo p. 34), recommendation (memo p. 37).
Links
FDA scan of the reviewer memo FDA briefing document for the February 8, 2001 meeting