Intended for healthcare professionals


Number needed to harm should be measured for treatments

BMJ 1998; 317 doi: (Published 10 October 1998) Cite this as: BMJ 1998;317:1014
  1. Arnold Zermansky, General practitioner
  1. Dib Lane Practice, Leeds LS8 3AY

    EDITOR—The concept of number needed to treat was an attempt to introduce both simplicity and objectivity into the evaluation of treatment. It is helpful both on a large scale when treatments are compared and care is commissioned and on the personal level when doctor and patient choose the management for the individual. The number needed to treat gives a comparison of treatments and outcomes that both doctors and patients should understand.

    One problem of this unimodal numerical scale for measuring the likely outcome of a treatment is that it oversimplifies the issues and may overvalue the treatment by ignoring its risks. Treatments may harm patients in various ways, the importance of which will depend on the disorder being treated as well as the nature of the harm. In a minor illness a potentially fatal treatment would not be acceptable even if the risk were fairly small. If a condition is uniformly fatal if untreated the risk of death or disability from the treatment is likely to be acceptable and a period of pain or discomfort may be a small price to pay. Not only are all adverse effects not equal but their importance depends as much on their context as their nature.

    Another problem with adverse effects is that we can seldom be as accurate in guessing risk as we are in measuring benefit. The therapeutic effect of a drug is usually unimodal and obvious. Controlled trials should separate the therapeutic effect from the placebo effect and allow an objective measurement of the real therapeutic benefit. One drug can, however, have many potential adverse effects, only some of which can be anticipated from its pharmacology or have been recognised when it is licensed. The incidence of common problems can be found from the results of clinical trials, but the recognition of more subtle effects often takes years. Even death and disability may not be recognised as therapeutic misadventure if the prevalence is small or the onset delayed. The risk will always be underestimated.

    We must not ignore the risk of treatment even if it is hard to measure. It may not be possible to devise a unimodal number needed to harm measurement, but a compensating negative measure is essential if we are not to delude ourselves and our patients about the value of treatment.

    A possible solution might be to separate adverse effects into several grades, on the basis of severity, reversibility, and usual duration. These might be: number needed to kill, number needed to disable, number needed to make you ill, and number needed to annoy. The concept of attaching a price list to the therapeutic menu should not stop with the cost of the pills.

    View Abstract