NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Rodgers M, Epstein D, Bojke L, et al. Etanercept, Infliximab and Adalimumab for the Treatment of Psoriatic Arthritis: A Systematic Review and Economic Evaluation. Southampton (UK): NIHR Journals Library; 2011 Feb. (Health Technology Assessment, No. 15.10.)

Cover of Etanercept, Infliximab and Adalimumab for the Treatment of Psoriatic Arthritis: A Systematic Review and Economic Evaluation

Etanercept, Infliximab and Adalimumab for the Treatment of Psoriatic Arthritis: A Systematic Review and Economic Evaluation.

Show details

Appendix 9Generalising the results of randomised controlled trials to general practice

Introduction

Chapter 3, Results of review of clinical effectiveness, showed that biologic drugs are much more effective than placebo controls in the experimental setting. The RCT is generally accepted as the best method to estimate an unbiased measure of the relative effectiveness of the treatment, in this case versus a placebo control, whether that relative effect is measured on a proportionate scale, such as an OR, or as a difference in means between groups. However, RCTs are not necessarily predictive of the absolute effectiveness of the intervention in general practice.

Any medical intervention can be thought of as a complex set of factors, of which the active pharmaceutical ingredients are only one component, albeit usually an important one. Other components of the intervention might include the relationship between the doctor and patient, interventions by other health professionals, and the patient's expectations, all of which to a greater or lesser extent, and for better or worse, contribute towards the overall outcome. Selection effects, or ‘regression to the mean’, may also play a part. These ‘non-pharmacological’ components of the intervention can be thought of as acting equally in the intervention and placebo arms of clinical trials, assuming that both doctors and patients are blinded as to the treatment arm. In these circumstances, the effect observed in the placebo arm of the trial measures the effectiveness of these non-pharmacological components, while the ‘treatment difference’ measures the independent effectiveness of the pharmacological component of the intervention.

Predicting the absolute effectiveness of the intervention in general practice requires some assumption to be made about whether the protocols, procedures and general ‘quality of care’ of the RCT are similar to general practice. A Cochrane Review218 found little evidence that using a placebo improved symptoms, with the exception of pain relief. However, the key question is not whether the ‘placebo effect’ is operating in every case, but whether outcomes associated with non-pharmacological components of the treatment are generalisable from RCTs to clinical practice. In other words, it matters less how the treatment works than whether it works.189

This generalisability would not matter too much if the decision model were comparing ‘placebo’ with ‘biologic therapy’, as both groups would experience the same non-pharmacological components of therapy. However, NICE will not compare an active therapy with a placebo, even if it were shown to be effective: it compares active therapies with ‘standard practice’ which in this case is assumed to be palliative care only. Adding the doctor's caring to the medical care component of biologic therapy might affect the patient's experience of treatment and may, for example, reduce pain and affect outcome. The ‘no-treatment’ group might or might not receive equivalent non-pharmacological care.

We can represent these possibilities as two scenarios:

  • Scenario 1 The ‘no-treatment group’ receives similar care (with similar mean outcomes) to the placebo arm in an RCT.
  • Scenario 2 The ‘no-treatment group’ receives less care than the placebo arm in an RCT, and does not achieve the response rate of the placebo arm in an RCT.

Conceptual framework

Figure 8 shows the mean change in HAQ ΔYjr from 0 to 12 weeks in the RCTs in the treatment group j = 1 and placebo group j = 0, depending on response, r = 1,0. These parameters were estimated in the evidence synthesis in Chapter 3. Variable α represents the change in HAQ over 3 months if there is no response for patients with placebo. Variable δ represents the mean difference in the change in HAQ between placebo non-responders and placebo responders. Variable βj represents the mean difference in the change in HAQ between placebo non-responders and non-responders with treatment, j. Variable γj represents the mean difference in the change in HAQ between placebo non-responders and responders with treatment, j.

FIGURE 8. Change in HAQ from 0 to 12 weeks in treatment groups estimated by RCTs.

FIGURE 8

Change in HAQ from 0 to 12 weeks in treatment groups estimated by RCTs.

The average change in HAQ (over responders and non-responders) in the placebo arm is:

ΔY0=[p0(α+δ0)+(1p0)α]=α+p0δ0

We can represent these scenarios by our beliefs about the relationship between the NH (i.e. the change in HAQ N in 3 months observed in general practice with no treatment) and the change in HAQ for non-responders in a placebo group (α), if both ‘placebo’ and ‘no treatment’ were compared in general practice.

Scenario 1. Results with ‘no treatment’ in practice are similar to placebo arms of randomised controlled trials

If N is approximately equal to α + p0δ (the average change in HAQ in the placebo group), this represents a scenario where we think the results obtained in a group given placebo, averaged across responders and non-responders, would be the same as what would have been observed if no treatment had been given.

In scenario 1, the absolute difference in the change in HAQ between treatment in practice and no treatment (difference-in-difference) can be estimated by substituting N = α + p0δ into the parameters shown in Figure 8 and so the difference-in-difference for responders is estimated to be (α + γj) − N + α + γj − (α + p0δ) = γjp0δ and for non-responders βjp0δ.

Scenario 2. The ‘no-treatment group’, in practice, gets worse outcomes than the placebo arm in an randomised controlled trial

In this scenario, patients with no treatment would not achieve the response rates observed in the placebo arms of RCTs. It is assumed that they would have the same outcomes as patients with ‘no response’ in the placebo group of an RCT. This implies that N is approximately equal to α. In this scenario, if placebo were to be given in practice, there would be some lasting average benefit over and above NH equal to: (α + p0δ) − N = α + p0δα = p0δ.

This might imply a lasting psychological benefit of the act of taking medication or could be due to beneficial interactions between the doctor and patient that occur both in trials and in the regular clinical setting. By extension, this ‘placebo effect’ would also partly explain the results in the treatment group, and would be expected equally in the trials and in general clinical practice. Therefore, we would expect that if biologic therapy and no treatment were compared in general practice, the absolute difference in the change in HAQ between treatment and no treatment (difference-in-difference) would be α + γjN = γj for responders and βj for non-responders.

It is difficult to test these alternative hypotheses, because the scenarios represent our hypothetical beliefs about a counterfactual argument: what would happen if ‘no treatment’, ‘placebo’ and ‘treatment’ were compared in general practice.

Conclusion

We conclude by setting out the implications for predicting the HAQ score in the decision model under each scenario.

In the decision model, variable N (the long-term NH in the untreated patients) is informed by observational evidence independent of the RCTs and is assumed to be constant over time. Therefore, in either scenario the HAQ score in the untreated group at time t after the start of the model is calculated as N × t.

If responders on treatment are assumed not to progress (worsen) over time, then the HAQ(t, j) score at time t for responders while still on treatment j is:

  • Scenario 1 Results with ‘no treatment’ are similar to average in placebo arms of RCTs (N = α + p0δ).
HAQ (t, j) = α + γj = Np0δ + γj
  • Scenario 2 The ‘no-treatment group’ achieves worse outcomes than the average in placebo arms of RCTs (N = α).
HAQ (t, j) = α + γj = N + γj

We assume that scenario 1 is the base case, consistent with the assumptions made in the previous Assessment Group model,177 and that scenario 2 is a sensitivity analysis.

© 2011, Crown Copyright.

Included under terms of UK Non-commercial Government License.

Bookshelf ID: NBK109471

Views

  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (1.7M)
  • Disable Glossary Links

Other titles in this collection

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...