Intended for healthcare professionals

Practice Diagnosis in General Practice

Diagnosis using “test of treatment”

BMJ 2009; 338 doi: https://doi.org/10.1136/bmj.b1312 (Published 24 April 2009) Cite this as: BMJ 2009;338:b1312
  1. Paul Glasziou, professor of evidence based medicine ,
  2. Peter Rose, university lecturer ,
  3. Carl Heneghan, senior clinical research fellow ,
  4. John Balla, visiting fellow
  1. 1Centre for Evidence Based Medicine, Department of Primary Health Care, University of Oxford, Oxford OX3 7LF
  1. Correspondence to: P Glasziou paul.glasziou{at}dphpc.ox.ac.uk

    Tests of treatment are commonly used when the diagnosis is uncertain, but can have pitfalls. The accompanying article (doi:10.1136/bmj.b1218) gives an example of how test of treatment can be used

    What is test of treatment?

    Though ideally we should have a clear diagnosis before starting treatment, such certainty is not always possible. Sometimes this uncertainty can be resolved by using the treatment as the test that confirms the diagnosis.1 For example, if we are unsure if a patient’s airway obstruction has a reversible element, a trial of steroids can test this: a sufficient response is then considered evidence of reversibility. At other times the test of treatment is not planned, but the failure to respond to treatment as expected leads to a rethink of the diagnosis. In this brief review we discuss different uses of the “test of treatment,” its reliability as a diagnostic tool, and how its use might be improved (for a specific example of its use, see the accompanying article on chronic cough2).

    When is it used?

    As illustrated in figure 1, a “test of treatment” is one strategy for the final stage of arriving at a diagnosis. It is appropriate when a single diagnosis is highly probable but not certain, when an available treatment works for most patients if the diagnosis is correct, and when there is a measurable short term outcome or surrogate. Such tests are more common and more useful in chronic or recurrent conditions rather than in acute conditions.

    Figure1

    Stages and strategies in arriving at a diagnosis

    A test of treatment is likely to be useful:

    • To make a diagnosis when the clinical features are atypical—for example, using glyceryl trinitrate for atypical chest pain or prednisone for suspected polymyalgia rheumatica in patients with a low or normal erythrocyte sedimentation rate

    • To make a definitive diagnosis when various differential diagnoses are possible—for example, chronic cough2

    • To see if a particular treatment—for example, switching antihypertensive drugs to minimise adverse effects—is appropriate in someone with the diagnosis.

    One early example of a test of treatment is the physostigmine test for myasthenia gravis,3 used since 1937. The treatment used for testing is not always the treatment used in the long term: short acting edrophonium (Tensilon) has become the standard agent to test for myasthenia gravis, although it is not suitable for long term treatment.4 Similarly with the trial of steroids (in several conditions): we usually won’t plan to continue long term oral steroids but will switch to other long term treatments. In the examples of tests of treatment in the table, the focus is a single condition, but sometimes a sequence of treatments can be used.

    Some common tests of treatment and their accuracy

    View this table:

    How does a test of treatment go wrong?

    As with every diagnostic test, the test of treatment can have both false negative and false positive results. If a test of treatment has been assessed against a diagnostic “gold standard” it is possible to quantify the accuracy of the test (table).1 For example, the combined results of five randomised trials of using proton pump inhibitors to “test” for gastro-esophageal reflux showed an average sensitivity of 78% and a specificity of 54% in comparison to 24 hour monitoring of pH.6 However, such accuracy has not been assessed for most tests of treatment.

    How can we improve?

    Since tests of treatment can easily lead to an inappropriate diagnosis, assessment of response to treatment should be more rigorous than in treatments where diagnosis is “certain.” A test of treatment has several potentially remediable problems. False positives can arise because of spontaneous remission of the condition or from placebo effects. False negatives can arise with an insufficient dose or duration of treatment, or if the patient is resistant to the particular treatment. Several things can reduce the influence of chance fluctuations in the condition or bias in evaluating response to treatment:

    • Use multiple measurements—if the measurements for an individual vary from occasion to occasion, getting several pretreatment measurements is helpful

    • Use multiple treatment periods—sometimes it is necessary and reasonable to withdraw and then reintroduce the treatment to provide convincing evidence of response. Using crossover periods (no treatment, treatment, no treatment, treatment, etc) provides more solid evidence; formally, this is the “n of 1” trial. For example, Guyatt reported that a patient thought their asthma was worse when they were taking theophylline. An n of 1 trial with six periods (three of theophylline, three of placebo) verified the patient’s impression. Also, for polymyalgia rheumatica some doctors use the “steroid sandwich”: patients keep a diary of symptoms for 21 days, during which they are given seven days of three tablets of Vitamin C, seven days of three tablets of non-enteric coated prednisone 5 mg, then seven days of three tablets of Vitamin C (the tablets resembled each other as nearly as possible). Sometimes such replication can be done in parallel—for example, in a test of treatment for yellow nail syndrome, Vitamin E was put on some of a patient’s fingernails, while that patient’s other fingernails were used as “controls”.12

    • Use blinded measurements or supplementary objective measurements—since we all often see what we hope to see, either blinded or objective assessment measures are needed to reduce bias.

    • Use objective measurements—if you don’t know what response you are looking for at the outset then it will be difficult to ascertain whether the patient has responded to the treatment at all. For example, a man with prostate obstruction may report that he has not improved, but by using the international prostate symptom score you can show him whether he has responded.

    In clinical practice, tests of treatment, whether formal or informal, are commonly used. In a sense every time we use a treatment, we are testing whether it works in a given patient. However, a false response may also lead to an incorrect diagnosis. By recognising and researching tests of treatment as a legitimate diagnostic test, we will be able to improve the accuracy and safety of their use.

    Learning points

    • Tests of treatment are common but imperfect

    • Clinicians should be aware of the accuracy of different tests of treatment

    • The accuracy may be improved by use of better measurements or more treatment crossovers

    Notes

    Cite this as: BMJ 2009;338:b1312

    Footnotes

    • This series aims to set out a diagnostic strategy and illustrate its application with a case. The series advisers are Kevin Barraclough, general practitioner, Painswick, and research fellow in community based medicine, University of Bristol; Paul Glasziou, senior clinical research fellow and lecturer in medical statistics, Department of Primary Health Care, University of Oxford; and Peter Rose, university lecturer, Department of Primary Health Care, University of Oxford

    • Thanks to Kevin Barraclough for suggestions on polymyalgia rheumatica.

    • Contributors: PG conceived the article; all authors assisted with the examples and writing and approved the final version. PG is guarantor.

    • Competing interests: None declared.

    • Provenance and peer review: Commissioned; externally peer reviewed.

    References