Rapid Responses to:

EDITORIALS: Alan Maynard Transparency in health technology assessments BMJ 2007; 334: 594-595 [Full text]

Rapid Responses: Submit a response to this article

Rapid Responses published:

.. not far enough

Jack Dowie   (26 March 2007)

NICE's model construction and assumptions should be explicit

Tom Fahey   (4 April 2007)

.. not far enough 26 March 2007
Jack Dowie, Emeritus Professor of Health Impact Analysis LSHTM WC1E 7HT Send response to journal: Re: .. not far enough	NICE needs to go much further. The whole cost-effectiveness modelling process needs to be opened up to involvement by all stakeholders, and inspection by the public, as it happens. The independent group should be contracted to develop the one and only model that NICE will consider in its appraisal, but do so in full and continuous collaboration with all registered stakeholders and interested parties who sign up to the NICE guidelines. The model will be run with any alternative parameters suggested by various parties and the alternative results (along with their evidential basis) submitted to the Appraisal Committee in a single report. No subsequent questioning of, or submissions regarding, the model structure, the results, or the underlying principles, would be accepted in the context of any particular case. Competing interests: None declared
NICE's model construction and assumptions should be explicit 4 April 2007
Tom Fahey, Professor of General Practice & Family Medicine, Royal College of Surgeons in Ireland Medical School, Mercers Medical Centre, Dublin 2 Send response to journal: Re: NICE's model construction and assumptions should be explicit	Editor, Alan Maynard raises important issues about the transparency of health technology assessment from NICE.[1] The methodological process that NICE uses is Markov modelling, a form of decision analysis that models the effectiveness and cost effectiveness of drugs or other medical interventions in a cohort of patients over time, producing incremental cost-effectiveness ratios for alternative strategies.[2] Construction of a Markov model relies on judgments concerning the likely outcomes of drug therapy- including benefits and harms. For this reason it is critical that full and transparent disclosure of Markov models occurs, facilitating open assessment of modelling assumptions and other methodological standards.[3] I was asked to comment on the updated hypertension guidelines produced by NICE and the British Hypertension Society (BHS) on behalf of the Royal College of General Practitioners as part of the consultation process in spring 2006. Drug therapy for hypertension is a controversial area as guideline recommendations between the US and UK differs in relation to the interpretation of recent randomised trials comparing newer and older antihypertensive agents. US guidelines recommend thiazide diuretics as first line agents based on evidence from the ALLHAT study, whilst NICE recommendations are based on a Markov model that includes evidence from ALLHAT and four other industry-sponsored RCTs. US commentators attribute differences in the results of these studies to inclusion of an inferior drug- beta-blocker Atenolol, as the initial comparator in three out of four of these RCTs and not including heart failure as part of the primary composite outcome.[4] My comments related to the transparency of the Markov assumptions including preferential modelling of diabetes as an adverse health state; failure to model differential withdrawals in different antihypertensive agents; and lack of probabilistic sampling in relation to effect size estimates for different classes of anti-hypertensive agents. Unfortunately, none of these suggestions were addressed in the published report and no comments on how the modelling and clinical guidance changed in relation to the consultation process is made. This is in marked contrast to the methods adopted by the Scottish Intercollegiate Guidelines Network (SIGN) that host an open meeting and incorporate consultation and peer review comments in an explicit manner (www.sign.ac.uk). If the academic community and the public are to have any confidence in NICE’s recommendations then transparency is required at all stages of the guideline development process. As Maynard states, knowing about the construction of the model and the underlying assumptions is a key part in this transparency.[1] Tom Fahey References 1. Maynard A. Transparency in health technology assessments: should NICE have the right to refuse access to its modelling data? BMJ 2007;334:594-5. 2. Littlejohns P, Kelly M. The changing face of NICE: the same but different. Lancet 2005;366:791-3. 3. Sonnenberg F, Beck JR. Markov models in Medical Decision Making: a practical guide. Medical Decision Making 1993;13:322-338. 4. Psaty B, Weiss NS, Furberg CD. Recent trials in hypertension: compelling science or commercial spin? JAMA 2006;295:1704-6. Competing interests: None declared