Minor bleeding episodes according to international normalized ratio (INR). Number of reports per person-years of exposure for men receiving active warfarin and placebo aspirin and those receiving active warfarin and active aspirin, and derived rate per 1000 person-years.
MacCallum PK, Brennan PJ, Meade TW, for the Medical Research Council's General Practice Research Framework. Minimum Effective Intensity of Oral Anticoagulant Therapy in Primary Prevention of Coronary Heart Disease. Arch Intern Med. 2000;160(16):2462–2468. doi:10.1001/archinte.160.16.2462
There is mounting evidence that low-intensity oral anticoagulation is effective, particularly in primary prevention of thrombosis, with important implications for safety and the practicalities of using warfarin. Because it is desirable to know possible benefits for different indications so that optimal therapy can be administered in as wide a range of conditions as possible, we analyzed data from the Thrombosis Prevention Trial, a factorial trial that compared treatment with low-intensity, dose-adjusted warfarin and low-dose aspirin separately and together, to determine the minimum effective intensity of oral anticoagulation in the primary prevention of coronary heart disease.
The international normalized ratio (INR) most recent to an event and overall time at each INR were used to calculate the INR-related event rate for coronary events, strokes, and major and minor bleeding episodes in 2545 men receiving warfarin with or without aspirin (75 mg/d) and followed up for a total of 9952 person-years.
Compared with placebo, warfarin alone at a dose that maintained the INR at 1.4 or more significantly reduced the risk of a coronary event by 47% (95% confidence interval, 4%-70%; P = .03), whereas the risk of a coronary event was not reduced at INRs below 1.4. Coronary events, strokes, and major bleeding episodes combined were significantly reduced by 45% (95% confidence interval, 9%-67%; P = .02) in the warfarin group compared with the placebo group when the INR was 1.4 or more. Minor bleeding episodes increased as the INR rose above about 2.0. No significant association of INR with coronary events was observed with combined warfarin and aspirin, possibly reflecting the small number of such events that occurred in this group, therefore limiting the power to detect an association with INR.
Warfarin alone is effective in the primary prevention of coronary heart disease when the dose is adjusted to maintain an INR of 1.4 or more. The results add to the evidence that low-intensity, dose-adjusted oral anticoagulation is effective for a range of conditions.
THERE IS mounting evidence that low-intensity oral anticoagulation is effective, particularly in primary prevention,1 with important implications for safety and the practicalities of using warfarin. It is therefore desirable to determine its possible benefit for different indications so that optimal therapy can be administered in as wide a range of conditions as possible. Because of the variability of the effect of anticoagulant therapy, it is difficult to study the optimal intensity of anticoagulation in a randomized manner2; therefore, observational studies (both follow-up and case-control) are necessary. Results of these studies have shown different optimal ranges in different conditions.3
The minimum effective intensity of oral anticoagulation in the primary prevention of coronary heart disease (CHD) is unknown; therefore, we sought to define this by using data from the Thrombosis Prevention Trial (TPT)4 to analyze the incidence of coronary events according to the achieved international normalized ratio (INR). The TPT was a randomized controlled trial with a factorial design that compared the effectiveness of treatment with low-intensity warfarin and low-dose aspirin in the primary prevention of CHD. Low-intensity oral anticoagulant therapy significantly reduced the risk of a first coronary event in middle-aged men at high risk. Analyzed on an intention-to-treat basis, warfarin adjusted to a target INR of 1.5 significantly reduced the risk of all coronary events (fatal and nonfatal) by 21%, fatal events by 39%, and mortality from all causes by 17%. The benefit of warfarin administered in this manner was at least as great as that seen with 75 mg/d of aspirin, while the combination of warfarin and aspirin reduced all coronary events (fatal or nonfatal) by 34%, ie, by more than either agent alone. These intention-to-treat analyses included the 50% of men who withdrew from treatment before completion of the trial and time at low INRs in those who remained "on treatment," so the value of warfarin above a minimum intensity while still receiving treatment might well be greater. The main side effect of warfarin therapy is bleeding, which in TPT was mainly confined to less serious events apart from combined treatment with warfarin and aspirin in men with elevated blood pressures who experienced an increase in cerebral hemorrhage. Although the number of these events was small, even taking them into account, combined warfarin and aspirin therapy prevented 12 coronary events for every stroke that may have been caused.
In addition to the relation between INR and coronary events, we also show results for these events together with strokes and major bleeding episodes to assess the overall balance between benefits and hazards in men receiving low-intensity warfarin alone or in combination with aspirin.
Full details of the trial have already been described.4 The trial was carried out through 108 group practices in the Medical Research Council's General Practice Research Framework in men aged between 45 and 69 years. At screening, attended by 61,422 (66%) of the 93,446 invited, smoking history and family history of premature CHD were elicited, body mass index calculated, blood pressure measured, and blood samples taken to test for total cholesterol and plasma fibrinogen concentrations and plasma factor VII coagulant activity. These variables were weighted according to their association with CHD in the Northwick Park Heart Study5,6 apart from family history, which was not recorded in this study and which for TPT was considered to increase risk by 50%. Within each practice, men in the top 20% of the resulting risk score distribution were considered to be at increased risk and eligible for the trial. Of the 10,557 men considered to be at high risk and eligible for the treatment phase, 5499 (52%) entered the trial, which started as a comparison of warfarin with placebo. Subsequently, all but 414 men entered a factorial comparison of warfarin and aspirin.
The factorial component of the trial, which was double-blind and placebo-controlled, resulted in the allocation of the 5085 men to 1 of 4 treatment groups: (1) active warfarin and placebo aspirin (warfarin) (n = 1268), (2) active aspirin and placebo warfarin (aspirin) (n = 1268), (3) active warfarin and active aspirin (warfarin-aspirin) (n = 1277), and (4) placebo warfarin and placebo aspirin (placebo) (n = 1272). The trial had approval from local research ethics committees and informed consent was obtained from all participants.
Warfarin sodium therapy was started at 2.5 mg/d and adjusted monthly until the target INR of about 1.5 was achieved when monitoring was undertaken every 3 months. Men were considered to be stable when 3 INRs of about 1.5 had been achieved. The median interval between starting warfarin therapy and stabilization was 5 months. INRs below about 1.3 were usually followed by a dose increase, whereas INRs greater than 1.8 led to a dose reduction that was implemented as an emergency if the INR exceeded 2.5. Treatment was interrupted during intercurrent illness, for surgery or dental work, when medication incompatible with warfarin was required, or when alcohol consumption was increased. Treatment was also deferred or interrupted in those whose blood pressure exceeded 170/100 mm Hg.
The primary end point was all CHD, defined as the sum of fatal and nonfatal events (ie, coronary death and fatal and nonfatal myocardial infarction). Fatal events were defined as coronary death and fatal myocardial infarction (death within 1 month). Stroke was a secondary end point, with thrombotic and hemorrhagic events determined by imaging or necropsy findings.
Records of the trial participants were flagged in the National Health Service Central Register to ensure automatic notification of the date and cause of death. The men visited the practice every 3 months to obtain new supplies of trial medication, to have their blood pressure monitored, and to complete a questionnaire on bleeding episodes. They were evaluated by their general practitioners every year, with a repeat electrocardiogram, and a research nurse annually checked the case notes for possible end points. The number for whom no information on possible nonfatal events was available was 58 (1.1%). Further inquiries about all possible events were made through the participating practices, hospitals attended, and, for fatal events, through coroners for all available clinical, investigational, and pathological evidence. An independent reviewer, unaware of treatment group, assigned events to the appropriate category (coronary death, fatal or nonfatal myocardial infarction [definite or possible], or stroke [thrombotic, hemorrhagic, subarachnoid, or of unknown type]) based on World Health Organization criteria.7
Information about bleeding episodes was obtained and classified under the following headings: major episodes—confirmed cerebral hemorrhages and fatal or life-threatening hemorrhages at other sites that required transfusion or surgery; intermediate episodes—determined separately from the routine questionnaires, usually by telephone, because they were more noticeable than minor episodes, including macroscopic hematuria, larger bruises, and prolonged nose bleeds; and minor episodes—determined by questionnaires completed each time the men attended their general practices during the course of the study, including questions about possible bleeding episodes (ie, bruising, nose bleeds, rectal bleeding, and pink or red urine).
Since it is the association of INRs with events that is under consideration, the data were analyzed on an on-treatment basis in this report, as in others dealing with the same question.8,9 The relation between INR and events has been established based on the method of Rosendaal et al2 to take account of the overall time at each INR and the INR most recent to an event. Because of the intervals between INR measurements in the main trial, we included in the analysis intervals between visits of up to 120 days (in the majority of instances this was 90 days) rather than the 56 days used by others.8,9 Each event (the numerator) was associated with the most recent INR measurement, providing this was obtained within the preceding 60 days of the event, ie, half the maximum time between consecutive INR determinations. To determine the time at each INR (the denominator), half the time from one INR to the previous measurement and half the time to the subsequent measurement was assigned as the period when that INR measurement was in effect. Data from the warfarin and warfarin-aspirin groups are shown separately to demonstrate the modifying effect of aspirin (if any). To ensure consistency and for comparison, the same criteria were used to calculate event rates in the aspirin and placebo groups, ie, events were only included if they occurred within 60 days of an INR measurement, and time on treatment was included up to a maximum of 120 days between INR measurements. Event rates (95% confidence intervals [CIs]) and event rate ratios were derived on the assumption of a Poisson distribution of events. Event rates were categorized into those that occurred at INRs less than 1.4 and those that occurred at INRs of 1.4 or more, this level of INR approximating the lower end of the trial's target range and dividing the overall time receiving warfarin into 2 roughly equal periods. Brief periods of interruption of warfarin therapy, eg, for dental work, were not included in the analysis. To allow for any differences between the on-treatment groups as a result of withdrawals from trial treatment, a multivariate Poisson regression model was used to control for confounding due to any ensuing effects of age, systolic blood pressure, smoking, cholesterol levels, fibrinogen levels, and aspirin use on event rates.
In the on-treatment analysis, 2545 men receiving warfarin were followed up for a total of 9952 years (mean per patient, 3.9 years), representing 62% of the total time considered in the intention-to-treat analysis. A total of 60,545 INR measurements were performed during the trial in the men receiving warfarin. The INR was within the target range of 1.3 to 1.7 for 65% of the total person-years, for 27% of the time it was less than 1.3, and for 8% of the time it was greater than 1.7.
While they were receiving warfarin treatment, participants experienced a total of 65 acute myocardial infarctions (15 fatal, 50 nonfatal) up to 60 days following the last INR, ie, 6.5 per 1000 person-years. There were 20 major, 81 intermediate, and 3515 minor hemorrhagic events during the treatment period, giving rates of 2.0 per 1000 person-years for major, 8.1 per 1000 person-years for intermediate, and 353 per 1000 person-years for minor bleeding episodes. There were 20 strokes during this period, of which 5 were thrombotic, 9 hemorrhagic (including 5 subarachnoid hemorrhages), and 6 of uncertain cause. The mean (SD) interval between most recent INR measurement and a thrombotic or bleeding event was 27 (19) days.
Withdrawals from the trial were almost equally divided among the treatment groups, which may have limited any possible bias in this on-treatment analysis and has also been allowed for by the regression analysis. On-treatment coronary event rates were lower than in the intention-to-treat analysis (Table 1). This probably reflects in part (1) the better outcome in trials of patients continuing treatment compared with trials including patients who withdraw, and (2) the method we used to calculate the on-treatment coronary event rate, ie, events were only included if they occurred within 60 days of an INR measurement. However, the on-treatment reductions with active treatment relative to placebo treatment, expressed as rate ratios, were very similar to the intention-to-treat benefits with the possible exception of a greater on-treatment reduction with warfarin-aspirin. Compared with placebo, aspirin reduced coronary events by 23% (95% CI, −17% to 49%; P = .27), warfarin by 18% (95% CI, −24% to 46%; P = .40), and warfarin-aspirin by 47% (95% CI, 15%-67%; P = .01). The presence of either active agent did not appear to reduce the beneficial effect of the other. Indeed, the reduction in coronary events with warfarin-aspirin was somewhat greater than would have been expected from a multiplicative reduction by the 2 active agents. Additional support for such an effect may come from the on-treatment data on minor bleeding episodes in the 4 treatment groups (Table 2). These data show similar increases in bleeding episodes with aspirin and warfarin compared with placebo, but an increase with warfarin-aspirin that is greater than the combined individual effects on bleeding of aspirin and warfarin compared with placebo (P for interaction, <.001).
The main results are shown in Table 3. Because there was an interaction of aspirin on the association of INR with coronary events (P = .03), the warfarin and warfarin-aspirin groups were examined separately. In the warfarin group, there was a 47% reduction in coronary events with INRs of 1.4 or more compared with placebo (95% CI, 5%-70%; P = .04). In contrast, achieved INRs below 1.4 were not associated with a reduction in coronary events compared with placebo. Indeed, there was a slightly higher coronary event rate in the former group. No significant difference in coronary event rate was observed between those with INRs of 1.4 or more and INRs below 1.4 in the warfarin-aspirin group and, if anything, the event rate was higher in the former group, although not significantly (P = .29). However, there were fewer events for this analysis in the warfarin-aspirin group compared with warfarin group (25 vs 40) and therefore the power to detect an INR-dependent effect is less. Of the time at an INR of 1.4 or more in either treatment group, 98% was spent in the INR range of 1.4 to 2.0.
There was no significant association between INR and strokes or major bleeding episodes in either the warfarin or warfarin-aspirin treatment group, although the absolute number of such events was small (Table 3).
In the warfarin group, an INR of 1.4 or greater was associated with a 45% reduction for the combined end point of all major adverse outcomes, ie, coronary events, strokes, and major bleeding episodes, compared with placebo (95% CI, 10%-67%; P = .02). There was no significant difference in combined major adverse outcomes compared with placebo when the INR was less than 1.4. With warfarin-aspirin there was no significant difference in the overall event rate with INRs of 1.4 or more compared with INRs below 1.4 (Table 3).
In the multivariate analysis, warfarin at an INR of 1.4 or greater reduced coronary events by 47% (95% CI, 4%-70%; P = .03), compared with placebo, whereas at INRs below 1.4 there was a nonsignificant increase (Table 4). With warfarin-aspirin, there was no significant difference in coronary event rates between those with INRs of 1.4 or more or INRs below 1.4. For the combined major end points of coronary events, strokes, and major bleeding episodes, warfarin was associated with a 45% reduction (95% CI, 9%-67%; P = .02) when the INR was 1.4 or more, compared with placebo.
There was no increase in the risk of minor bleeding with increasing INR until a level of approximately 2.0, when the risk increased significantly (P<.001) (Figure 1). The addition of aspirin to warfarin was associated with roughly a doubling in the risk of minor bleeding compared with warfarin alone (P<.001), and when the INR rose to 2.2 the bleeding risk with warfarin became equal to that of warfarin-aspirin, with an INR of 1.8 or less.
Usually, on-treatment analyses show larger benefits than intention-to-treat analyses because inclusion in the latter of those who have withdrawn from trial treatment inevitably dilutes the true pharmacological effect. It is therefore surprising that the reductions (rate ratios) in the 2 types of analyses are similar for aspirin alone and warfarin alone (Table 1). For combined treatment of warfarin-aspirin, the on-treatment analysis suggests a larger reduction, but the difference compared with intention to treat is not statistically significant. If there is indeed no difference in the proportional reduction between the 2 analyses, there may be longer-term beneficial effects of both aspirin and warfarin than the short-term effects on platelet aggregability and fibrin formation usually considered. The possibility that both agents affect the development of chronic vessel wall pathology has been raised. Warfarin may reduce the amount and therefore the consequences of both thrombin and fibrin generation within the vessel wall,10,11 which in turn may outweigh the potential disadvantage of reduction by warfarin of matrix GLA protein, a vitamin K–dependent protein that is thought to protect the vessel wall against calcification.12 Platelets also contribute to atherosclerosis13 and by its antiplatelet effect, aspirin too might modify vessel wall pathology, although as yet there is little clinical or epidemiological evidence to support this.
By defining a lower INR limit to the therapeutic range, our on-treatment analysis adds to the information from the intention-to-treat analysis,4 a benefit of this methodological approach that has been noted in other studies of oral anticoagulation.3 In assessing the risk of clinical events at different INRs we have based the analysis on the method of Rosendaal et al,2 which takes account of the overall time at each INR. This method assumes a linear change in INR between 2 consecutive measurements which, although perhaps not accurate for an individual whose dose of warfarin is adjusted, is likely to even out within the context of a large observational group, particularly if anticoagulant control is fairly stable. Because of the routine interval between INR measurements in the trial, we allowed the attribution of events to an INR determined up to 60 days (mean interval, 27 days) before the event, compared with 28 days by others.8 Our analysis may therefore have underestimated the probable effect of higher INRs in reducing event rates because of the imprecision resulting (for any interval) from the association of events with INRs some days previously. However, the clear and expected association of progressively increasing rates of minor bleeding episodes as the INR increased above 2.0 has confirmed the general validity of our approach.
We arbitrarily categorized the event rates into those at INRs of 1.4 or more or below 1.4, because this level approximated the lower end of the target range and divided the overall time receiving warfarin into 2 roughly equal periods. Our results show that it is necessary to maintain the INR at 1.4 or more for warfarin alone to be effective in the primary prevention of CHD. The antithrombotic efficacy of low-intensity, dose-adjusted oral anticoagulation in the primary prevention of CHD is consistent with data showing benefit in other clinical conditions. On-treatment analysis of the Stroke Prevention in Atrial Fibrillation III trial14 in patients with nonrheumatic atrial fibrillation suggests lower event rates for stroke or systemic embolism with warfarin alone at achieved INRs between 1.5 and 1.9 compared with INRs less than 1.5 with warfarin in combination with aspirin. Observational data from a large cohort study in patients with a variety of indications for oral anticoagulant therapy15 and a recent case-control study in patients with atrial fibrillation16 show that INRs in the 1.5 to 2.0 range are associated with fewer thrombotic events than INRs less than 1.5, although the risk is further reduced as the INR rises to 2.0 or more. A randomized trial has also shown that lower-intensity anticoagulant therapy (warfarin, 1 mg/d for 6 weeks followed by dose-adjusted warfarin with a target INR of 1.3-1.9) is effective in prophylaxis against venous thrombosis in patients with stage IV breast cancer receiving chemotherapy.17
Because there was an interaction of aspirin on the association of INR with coronary events, reflecting the different directions of the associations of these events with INR in the warfarin and warfarin-aspirin groups, these treatment groups were examined separately. In contrast to the warfarin-alone group, we did not find a reduction in coronary events with increasing INR in the warfarin-aspirin group. Indeed, there was a slightly greater coronary event rate in those receiving warfarin-aspirin with INRs of 1.4 or more compared with INRs below 1.4, although this was not statistically significant. There are 3 possible explanations. First, the finding may reflect the small number of coronary events and the relative imprecision of the method we used to relate INRs to events, as noted above; in other words, the interaction, which is significant at only a marginal level and had not been prespecified, is due to chance. Second, aspirin and warfarin taken together might have antagonistic effects. However, this explanation can be ruled out with virtual certainty because of the greater clinical benefit of combined treatment both in TPT and, in other circumstances, studies in a growing number of trials using combined treatment.18,19 Third, and assuming that there is no difference between warfarin-aspirin treatment at INRs below 1.4 and at 1.4 or more, the intensity of oral anticoagulation with warfarin may be immaterial if aspirin is also being taken. Against this possibility are the results of other trials of combined therapy that have reported benefits when the INR has been adjusted18,19 but not when the dose of warfarin has been fixed or capped,14,20- 22 although these trials all involved patients at greater risk of a thrombotic event than was the case in TPT, and also involved a shorter follow-up. We are inclined to attribute the finding to chance. Whatever the explanation, the result in those taking warfarin alone is clear, and indicates that an INR of 1.4 or more is necessary to confer benefit.
In summary, our results show that at least for warfarin alone to be effective in the primary prevention of CHD, the achieved INR needs to be 1.4 or more. Prolongation of the INR to between 1.4 and 2.0 in high-risk men is associated also with a reduction in the overall risk of major adverse events, ie, coronary events, strokes, and major bleeding episodes combined. There is no evidence of an INR-related increase in the risk of bleeding until an INR of approximately 2.0 is reached and even then the risk is largely confined to minor bleeding. This information is helpful in defining a lower limit of the therapeutic range for warfarin, which may be applicable in other settings. Combined treatment with low-intensity, dose-adjusted warfarin and low-dose aspirin is effective in the primary prevention of CHD. However, since the addition of aspirin to warfarin is associated with an increase in the risk of bleeding, although this is mostly minor, it might be reasonable in practice to decide which single agent to use and confine combination therapy to situations where the risk of thrombosis is particularly high. If warfarin at an INR of 1.4 or more reduces coronary events by 47% and if (but not only) this benefit is mainly in preventing fatal coronary events,4 the use of low-intensity, dose-adjusted oral anticoagulation for primary prevention should be more widely considered, particularly since 25% to 30% of individuals die in their first major coronary episode. Given that administration of anticoagulant therapy (and aspirin) always requires a judgment to be made on the balance of risks of thrombosis and hemorrhage, the results of TPT together with those of other trials support the further study of low-intensity, dose-adjusted warfarin (target INR 1.5 to 2.0) in other conditions, with the important implications of less bleeding and reduced frequency of monitoring.
Accepted for publication May 19, 2000.
The trial was funded by the Medical Research Council, London, England, the British Heart Foundation, London, DuPont Pharma, Wilmington, Del, and Bayer Corporation, Leverkusen, Germany. DuPont Pharma provided warfarin and Bayer Corporation provided aspirin free of charge during the main trial.
Corresponding author: Peter K. MacCallum, MD, MRCPath, MRC Epidemiology and Medical Care Unit, Wolfson Institute of Preventive Medicine, St Bartholomew's and the Royal London School of Medicine and Dentistry, Charterhouse Square, London EC1M 6BQ, England.