38 Mono-criterial thresholds are a likely source of over-testing and over-treatment

Abstract

Objectives To establish that the optimal decision for the individual person, which is preference-sensitive and multi-criterial, can be identified without any mono-criterial diagnostic classification, such as one based on threshold segmentation of a risk or disease scale. These segmented subgroups may be needed in research to produce the probabilistic prognoses for each option required in person-centered decision making. However, their use in the decisional context will often lead to suboptimal decisions for the individual, leading to over-testing and over-treatment from their point of view. Such classifications also jeopardise the obtaining of fully informed and preference-based consent to subsequent actions. Method Personalised decision support tools based on the technique of Multi-Criteria Decision Analysis (MCDA) have been produced for many conditions, including various cancers and cardiovascular conditions, and for the screening, testing and treatment contexts. Some are original, most are translations of published studies into MCDA format. A person-centred MCDA-based decision support tool requires the performance of all options on all criteria to be established on a continuous 0–1 scale. (Network Meta-Analyses provide the preferred evidential basis for these performance rates, but expert opinion is almost always required for option-criteria cells that lack robust evidence. The person is themselves the expert on a criterion such as Treatment Burden). Any threshold placed on any one of the multiple criteria in the analysis interferes with the cross-criteria validity of a value-based, compensatory MCDA and hence jeopardises its overall integrity. Results A full set of option evaluation scores is produced by the personalised decision support tool without any reference to mono-criterial risk classes or disease stages, such as those based on PSA and Gleason score thresholds for prostate cancer, on euroSCORE thresholds in cardiovascular disease, and Frax fracture risk thresholds in bone health. The significant weights assigned to the other criteria, most of which are harms, ensure that the optimal decision for many individuals is not that based on the focal criterion, implicitly assumed to be solely or very highly weighted. As one specific example, in the piloting of a tool with patients having a first DXA bone scan (n=32), the mean weight to Avoiding a Fracture was 43%, with 37% assigned to Avoiding Side-effects, and 20% to Avoiding Treatment Burden. Conclusions Testing and/or treating based on segmentation of a continuously-scaled criterion ignores the other criteria analytically, leaving them to be ‘taken into account’. It also undervalues the other criteria important to the person, most of which are harms. The segmented mono-criterion is almost always benefit-focused, so the excessive weight given to it, relative to these harms, logically leads to over-testing and over-treatment for that person. The multi-criterial decision support tools confirming this pattern have been evaluated by an embedded multi-criterial Decision Quality PRO, but here we make empirical claims only at the proof-of-method level. (If the person asks to be provided with a label (e.g. ‘high’, ‘moderate’ or ‘low’ risk) this request should be addressed ex post engagement with the tool and they should be offered support in a personalised assessment of the benefits and harms of labeling.