Consent for the use of personal medical data in research

Web Extra

[posted as supplied by the authors.]

Different Consent Processes and associated Costs

If we compare four possible consent processes (purely in terms of cost):

• ‘Express Consent’ – where each potential participant is individually informed about the study, and their consent sought (and possibly formalised on paper)
• ‘Opt-out’ – where each potential participant is individually informed about the study, and included unless they object;
• ‘Social Contract’ – people are informed generally about research and information is freely available on specific projects and will be included unless they object (either generally to being a research subject or to being included in a specific study) – basically the costs of information provision are greatly reduced.
• ‘Anonymise’ – where data is de-identified, possibly leaving a pseudonym for linking and tracing if needed, so that there is no legal requirement for consent. However, this is still a point of contention and no clear guidelines exist. We presume that this is a viable option without impacting the results of the study.

For any study, the statistical power of the study will depend on the number of cases or participants considered. From the law of Large Numbers, the accuracy of results will increase with the square-root of the number of cases (reducing the standard deviation of the sample mean). If we make some assumptions about the likely benefits of a study, then we could calculate the likely benefit per case as a diminishing quantity (the overall benefit increases though, but at a lower rate than the number of cases) shown as a dark blue band in fig 1 in the main article, using a log-log scale.

If we used an opt-in approach, then the cost per case would be constant (though scaled up by the recruitment ratio, e.g. that five times as many people need to be approached in order to get the required number of participants) – this appears as flat blue line in fig 1. While the example uses arbitrary but reasonable values, it is inevitable that the grey band and the blue line intersect at some point (between 10,000 and 100,000 in the diagram) so that there is some upper bound where there is no point recruiting further candidates. Note that is rather assumes that the actual cost of the study, analysing the data, is comparatively negligible.

If we apply an opt-out approach, then we will have to fund a public information campaign (so that potential objectors are aware of the study, their options, and how to opt-out from the study) and a per case cost of managing the objections. This is shown in the diagram as the upper light-blue line, which drops faster than the benefits band as the campaign cost is defrayed over more cases. For smaller numbers, this is simply not cost-effective; for larger numbers, it may well be, provided the percentage of objections is not too great (as the curve inevitably flattens out) – in this case, above about 300,000.

The third approach considered is a ‘social contract’ whereby a general information campaign for research rather than one specifically for each study is used, though there would have to be some specific provision for each study, so that potential objectors could discover what research might affect them and register their objections. In the diagram, we have assumed that the public information costs are reduced to, say, 20% of the full campaign assumed in the ‘opt-out’ costs. Even here matters are getting tentative at 10 million cases.

The final approach is anonymisation. It is assumed that some anonymisation procedures and routines need to be developed, which may include getting these vetted and approved whatever the size of the study. This is effectively what happens when projects apply to the Patient Information Advisory Group (PIAG) to gain exemption from common law requirements under Section 60 of the Health & Social Care Act 2001.

Basis of calculations

‘Express Consent’ cost – taken as £50 per case, covering cost of locating candidate, making contact (usually indirectly through their GP – recommended BMA cost of £108 per hour), providing relevant information, either in person or via posted leaflets and supported web-site for more details), recording consent; geared up by recruitment percentage (so with only 20% recruitment costs per case recruited are five times that per candidate actually approached). Wilson et al (BMJ. 1999 May 29; 318(7196): 1484) also suggest £50 (though at 1999 prices) – Ward et al (BMJ 2004;329:277-279) quote costs of £300 per case (but £1,100 for a control). The cost per case is then simply a flat £50

For the ‘opt-out’ approach, it was assumed that a public information campaign would cost, say, £10 million, though this might be modified if the study were in some way geographically limited to a region, but multiple media and channels would be needed to ensure a reasonable coverage. We assumed a 5% opt-out rate with each case incurring a similar cost to gaining express consent. Clearly a higher opt-out rate will increase the costs. The cost per case will be £10,000,000/N + 5% x £50

For the ‘Social Contract’ model, it was assumed that the public information campaign costs would be reduced to £2 million, using perhaps just leaflets at relevant clinics as well as a public web-site. The cost per case will be £2,000,000/N + 5% x £50

For ‘Anonymise’, we have assumed that this will be perhaps £250,000, though clearly for a smaller study more direct manual approaches might be more cost-effective. There will be a small per case cost of anonymising and maintaining any pseudonym, taken to be about £2. Ideally, systems and techniques for anonymising or de-identifying data should be shared, which would lower the development costs.

For benefits, then much will depend on the outcome of a study, positive or negative and whether clear action could be taken to improve health, reduce the cost of care, improve the efficacy of care, or whatever. However, on average, one might assume that benefits for a 1,000-person study might be perhaps £150,000 to £450,000 - the formulae were taken to be £5,000 x √N/N (lower-bound) and £15,000 x √N/N (upper-bound).

Changing some of these figures makes relatively little difference to the overall implications, though it can change some of the points of transition: trebling consent costs to £150, halving the public information costs, and increasing the benefits (using £10k and £25k in the formulae) gives the following graph:

Figure 2: Higher benefits and costs

Express consent is still viable for smaller studies, opt-out approaches do become more viable, particularly the ‘social contract’ approach, but for a full population model of over 10 million cases benefits are only just meeting costs. If we increase the opt-out to 10%, then in fact not even the ‘social contract’ model is viable:

Figure 3: 10% opt-out effects

This would suggest that we need a ‘no consent’ approach for such databases justified by strong security and confidentiality processes which limit risk by having ‘effectively anonymised’ data and controls on access for specific projects only (possibly providing aggregate data only) – the ‘anonymise’ option, but this needs clear support in law or formal guidance. This is the model for the Secondary Uses Service (SUS) under the National Programme for IT, though it is not yet established how researchers may gain access as it is currently intended for NHS staff only. This is discussed in the AMS report [3]