How a single adverse event can affect future prescribing: matched pair analysis

A single adverse event can cause doctors to forget the evidence base. Tom Treasure looks at a matched pair analysis that assessed future prescribing of warfarin in doctors whose patients with atrial fibrillation had haemorrhage or thromboembolic stroke

This month's paper is Choudhry NK, Anderson GM, Laupacis A, Ross-Degnan D, Norman SLT, Soumerai S. Impact of adverse events on prescribing warfarin in patients with atrial fibrillation: matched pair analysis. BMJ 2006;332:141-3. You can read it by going to studentbmj.com and clicking on the link.

Abstract
Objectives—To quantify the influence of physicians' experiences of adverse events in patients with atrial fibrillation who were taking warfarin.Design—Population based, matched pair before and after analysis.Setting—Database study in Ontario, Canada.Participants—The physicians of patients with atrial fibrillation admitted to hospital for adverse events (major haemorrhage while taking warfarin and thromboembolic strokes while not taking warfarin). Pairs of other patients with atrial fibrillation treated by the same physicians were selected.Main outcome measures—Odds of receiving warfarin by matched pairs of a given physician's patients (one treated after and one treated before the event) were compared, with adjustment for stroke and bleeding risk factors that might also influence warfarin use. The odds of prescriptions for angiotensin converting enzyme inhibitors before and after the event was assessed as a neutral control.Results—For the 530 physicians who had a patient with an adverse bleeding event (exposure) and who treated other patients with atrial fibrillation during the 90 days before and the 90 days after the exposure, the odds of prescribing warfarin was 21% lower for patients after the exposure (adjusted odds ratio 0.79, 95% confidence interval 0.62 to 1.00). Greater reductions in warfarin prescribing were found in analyses with patients for whom more time had elapsed between the physician's exposure and the patient's treatment. There were no significant changes in warfarin prescribing after a physician had a patient who had a stroke while not on warfarin or in the prescribing of angiotensin converting enzyme inhibitors by physicians who had patients with either bleeding events or strokes.Conclusions—A physician's experience with bleeding events associated with warfarin can influence prescribing warfarin. Adverse events that are possibly associated with underuse of warfarin may not affect subsequent prescribing.
What's it all about

This paper deals with common but serious problems—atrial fibrillation, the consequent risk of stroke, stroke prophylaxis with warfarin, and bleeding related to taking warfarin. Evidence was accumulating in the 1990s that lone atrial fibrillation (not associated with valve disease or a structural abnormality of the heart) is responsible for 15% of all strokes. Anticoagulation with warfarin reduces the risk of stroke. Whenever drugs are used to reduce the coagulability of the blood, we balance the risk of the thrombus formation, which we are trying to prevent, with the risk of bleeding, which we induce with the treatment. In the case of lone atrial fibrillation, despite the risk of bleeding caused by warfarin, the combined risk of stroke and bleeding is lower. But doctors do not give anticoagulant drugs to between a third and two thirds of their patients with atrial fibrillation despite good evidence of benefit.

Why I chose this paper

This study attracted my attention for a personal reason. About 10 years ago, I advised a screenwriter who had a story in which the husband was a heart surgeon and his wife a rural general practitioner. The screenwriter wanted to make a contrast between their worlds; my task was to suggest an authentic illustration. While the general practitioner was visiting a favourite elderly patient, I suggested that she make the clinical diagnosis of atrial fibrillation and say something like, “I know the clever chaps would send you to the anticoagulant clinic but . . . sigh . . . perhaps we'll give you some digoxin and settle for that.” It worked in the drama, the contrast was made, the doctor eschewed intervention in favour of compassion, but the decision was wrong. It was not the better course of action in the old man's overall interests. To be fair to myself and my fictional doctor, the evidence had not yet been published.¹

Questions

The researchers consider the following questions in their study.

Do doctors underprescribe warfarin in patients with atrial fibrillation because their judgment is swayed by an experience of a patient bleeding after taking warfarin?

If a patient of theirs in atrial fibrillation has a stroke, how does that influence the doctor's behaviour?

How did the researchers try to answer these questions? Pause first to consider the difficulties. You might devise a questionnaire. Too often researchers resort to questionnaires, but, unless expertly designed, these are self serving. How people behave and how they would like to think they behave can differ widely.

Why not use a prospective study? Studies of behaviour offer the study participants the opportunity to change their behaviour—they know they are under scrutiny. Also, the rates of the events are low single figure percentages—data will take a long time to accrue.

The researchers used a database—the records of the Canadian Institute for Health Information—to identify 116 200 patients with lone atrial fibrillation. They identified two subgroups: 3921 patients had been treated with warfarin and had subsequently been admitted with gastrointestinal or intracerebral haemorrhage. Another 8720 patients had not been treated with warfarin but had been admitted with a stroke likely to be thromboembolic.

From another database (that of the Ontario Health Insurance Plan) the researchers identified the doctor most involved in each patient's care. Now comes the clever bit. The researchers shifted their attention from these patients to their doctors. Did the doctors' habits in prescribing warfarin change when a patient under their care had the serious clinical event of bleeding or stroke? The researchers compared the 90 day periods before and after this event.

Answers

The 530 doctors whose patients had bled were less likely to prescribe warfarin for atrial fibrillation afterwards—a fall from 48.5% to 41.9% (P=0.03). The 704 doctors whose patients had a stroke had similar prescribing rates for warfarin (36.9% v 35.9%). These results are unadjusted. The authors checked confounding factors by conditional logistic regression. This enabled them to adjust for other factors—whether the doctor was a specialist cardiologists or not, for example. This did not alter the conclusion.

Another important check was made: the researchers looked at the number of prescriptions for angiotensin converting enzyme inhibitors. This did not change, making extraneous variation in prescribing habits an unlikely explanation.

Conclusions

The experience of a single patient having a haemorrhage changed the prescribing practice of the doctors to the extent that they underused warfarin in subsequent patients. But the experience of a single patient having a stroke did not encourage them to prescribe anticoagulants more readily.

In an ideal evidence based world

After admitting a patient for bleeding, the doctor should have considered any special factors in that patient. But the benefit over the risk remains the same for all subsequent patients. This single event should not have changed evidence based practice. Bleeding after taking warfarin is a recognised and calculated risk of anticoagulation.

After an admission for stroke, an appropriate response may have been to review practice and note the evidence—stroke in a patient in atrial fibrillation may be averted by anticoagulation. This might have led to more prescriptions for warfarin in subsequent patients with atrial fibrillation.

As a doctor you can make errors of commission and errors of omission. In this instance, the doctors changed their practice contrary to the research evidence, if the treatment that they had prescribed had caused the complication. On the other hand, when patients had a stroke when they had not been prescribed anticoagulants the error of omission was disregarded.

All interventions include risk, which must be balanced against the risk of not acting. Examples are the vaccination debate and the programme of screening for abnormal pregnancies. A few children harmed by vaccination impinge more on public consciousness than the health of the many children spared disease by the vaccination programme. And inadvertently induced abortion in the course of amniocentesis is the price to be paid to spare mothers a severely abnormal fetus.

DANIEL SAMBRAUS/SPL

Methods and statistics

For many purposes, we prefer prospective data because we collect only the data that we are going to use, and we can be sure that we collect them as completely as possible. Data collected and stored without any prior hypothesis, however, have the advantage that they are fixed and cannot be distorted by the study question. If you were to prospectively study warfarin prescribing habits, they might well change in the course of the study. This study is a good example of the value of databases.

Are we confident with the statistical analysis? As readers, we have to largely trust the integrity of the presented data and the skill with which they were analysed. And we rely on vetting by statistical reviewers to make sure that we are not being sold a dud. Table 1 in the unabridged version contains 36 comparisons and it is more likely than not that two of them will be different at the 5% level. A P value of 0.05 means that 19 times in 20, the observed difference would not have occurred by chance. Because in this instance the prior hypothesis was based on warfarin use, it is not a chance finding and is likely to be real.

Although prescriptions of warfarin fell after a bleed, prescriptions for antiplatelet drugs were unchanged (including for the apparently safe low dose aspirin). But warfarin can be reversed more easily than platelet inactivation. This is a common simplification—that warfarin is dangerous and aspirin is safe. Non-steroidal anti-inflammatory drugs can cause bleeding, but they were not mentioned by the authors. Prescribing fell from 30% to 25% (P=0.07). This might well have been the doctors' choice rather than chance.

Sophisticated statistical analyses help to unravel these conundrums so that we can draw the best conclusions and avoid errors, but there is no harm in looking at the data that went into the analysis to check that it makes sense.


Tom Treasure, professor of cardiothoracic surgery, Guy's and St Thomas's Hospital, London SE1 9RJ
Email: Tom.Treasure@ukgateway.net


studentBMJ 2006;14:133 - 176 April ISSN 0966-6494

1. Hart RG, Benavente O, McBride R, Pearce LA. Antithrombotic therapy to prevent stroke in patients with atrial fibrillation: a meta-analysis. Ann Intern Med 1999;131:492-501.