Skip navigation
Link to Back issues listing | Back Issue Listing with content Index | Subject Index

Computer decision aids for anticoagulation

Review
Results
Comment
One of the benefits of using computers is that they remember things that we may forget. They can also remember numbers, and perform calculations, and should be better than we are at coming up with the right answer more often, if programmed correctly at the beginning. The trouble with people is that their programming and processing are both of variable quality.

One place to test the ability of computers to aid in therapeutic quality should be in anticoagulation, where the risks and benefits depend heavily on maintaining a narrow limit. Small improvements in quality control would be useful. Even better would be a randomised trial comparing computer with standard therapy, and even better than that would be a meta-analysis of such studies [1].

Review


The review [1] searched MEDLINE using some standard search terms. Papers were selected if assignment of patients to computer systems or not was randomised, and if there was information available relevant to the analysis. The outcomes were the number of tests within the target range for anticoagulation. Also assessed was any information on major haemorrhage or bleeding events.

Results


There were seven studies with 3416 anticoagulation tests. Computers generally did better (Figure 1). With computer systems 65% of tests were within target compared with 59% without computers. The number needed to treat was 17 (95% confidence interval 11 to 38), meaning that for every 17 patients whose anticoagulation was controlled by a computer system, one more would have a test result within target compared with not using a computer system. Other statistical outputs are in Table 1, including the odds ratio quoted in the paper.

Figure 1: Results of seven RCTs of computer assisted anticoagulation - percent of tests within target range for INR


Table 1: Different ways of describing the same result (better control with computer)

Output Result
Odds ratio 1.29 (1.12 to 1.49)
Relative risk 1.10 (1.04 to 1.16)
NNT 17 (11 to 38)
Percent in target with computer 65 (63 to 67)
Percent in target without computer 59 (57 to 62)
Percent bleeds with computer 2.0 (1.0 to 3.0)
Percent bleeds without computer 4.4 (2.8 to 6.0)

There were 14 major haemorrhages among 700 subjects in the computer groups (2.0%), versus 25 among 636 in the control groups (3.9%). The relative risk was 0.51 (0.27 to 0.97), and the number needed to treat to prevent one major haemorrhage was 52 (26 to 1013).

There were 25 major haemorrhages, deaths and thromboembolic events among 700 subjects in the computer groups (4.0%), versus 39 among 636 in the control groups (6.1%). The relative risk was 0.65 (0.40 to 1.04).

Comment


There are several interesting things about this review and meta-analysis. Firstly it identified seven randomised trials, compared with four (three common) found for another review of computerised systems [2]. The most likely reason for the difference was different inclusion criteria.

Second, it showed that computer-aided decisions resulted in a moderate improvement in quality, as well as reducing by half the number of major haemorrhages. The amount of information is still not great, so we must remain cautious. But adverse events can often be expensive for healthcare systems as well as tragic for patients and professionals. A back of envelope calculation would suggest that saving one major bleed for every 50 patients would make any computer system pay for itself.

Third, the paper showed how odds ratios can be wrong and misused. Here the odds ratio was 1.29, when the relative risk was 1.1. Some argue that odds ratios should not be used when proportions are high [3]. Even worse, the authors comment "the use of a computer for anticoagulation optimization increased by 29% the proportion of visits where patients were within the therapeutic range". Oh no it didn't: it went up from 59% to 65%, and increase of about 10%.

Misuse of odds ratios is common, and it is often wrongly taught. The real problems come when there is genuine disagreement about how and what to use in particular situations. Not everyone agrees about when odds ratios are right and relative risk wrong, or vice versa [4].

Odds ratios are not for us common folk.

"All policy decisions should be based on absolute measures of risk: relative risk is strictly for researchers only" [5]

If we stick to absolute numbers we can probably work it out for ourselves as long as we have a statistical tick.

References:

  1. G Chatellier et al. An overview of the effect of computer-assisted management of anticoagulant therapy on the quality of anticoagulation. International journal of Medical Informatics 1998 49: 311-320.
  2. R Walton et al. Computer support for determining drug dose: systematic review and meta-analysis. BMJ 1999 318: 984-990.
  3. DG Altman et al. Odds ratios should be avoided when events are common. BMJ 1998 317: 1318.
  4. S Senn. Rare distinction and common fallacy. www.bmj.com/cgi/eletters/317/7168/1318.
  5. Rose. J Roy Coll Phys Lond 1991 25: 48-52.
previous or next story in this issue