Formative preference-sensitive measures are needed in person-centered healthcare at both clinical and policy levels

Abstract

Background: Most existing multi-critierial instruments measuring the person-as-patient’s experience (PREMs like ‘decision quality’) or outcome (e.g., PROMs like ‘quality of life’) are designed for use as inputs into group level decision and guidelines. Regrettably, they elicit only the ratings of the individual respondent. They therefore lack person-centered preference-sensitivity in that the measure for the individual patient does not also reflect that patient’s relative weighting of the multiple items/criteria/indicators that constitute it. Some instruments may use weightings that are preference-based, but the preferences are almost always the average ‘tariff’ derived in a separate study, not those of the individual concerned. The aggregated results from such instruments cannot reflect the heterogeneous and potentially changing preferences of the patient population and in this sense are not the appropriate ‘living’ metrics.

Personalisation: A preference-sensitive person-centered instrument reflects these weightings and is ‘dually-personalised’. Six possible degrees of personalisation are identified, varying in the ability of the individual respondent to customize the instrument. The appropriate level depends on the context and goals of the measurement. Higher levels of personalisation, such as those allowing the individual to generate the items, are unlikely to be of use in routine clinical practice, where dual and triple personalisation are the most practical. (In triple personalisation the respondent selects their criteria from a menu.)

Constructs: Whether single or some higher degree of personalisation is appropriate depends on the nature of the construct. Formative constructs producing indexes, such as most PROMs/PREMS, do not exist independently of the instrument claiming to measure them. The multiple components of these instruments, such as pain and mobility, ‘cause’ the construct and are not caused by it. Measures of formative constructs should always be (at least) dually-personalised. In contrast, reflective constructs (such as Appendicitis), exist independently of measurement and the scale measure they produce should never be dually-personalised; the component cues are properly integrated by belief/evidential coefficients, not value judgments.

Validation and Measurement models: Currently, most PROMs and PREMs are being treated as reflective indexes and are only singly-personalised. In breach of the COSMIN guidelines they are often ‘validated’ by criteria such as internal consistency and structural validity, using techniques such as Rasch or Item Response Theory, that are appropriate only for reflective scales. The output from applying a conjoint measurement model to a preference-sensitive construct (such as quality of life) is an appropriate input only in single-criterion decision-making, where the measured construct is the sole maximand. It is not appropriate in multi-criteria decision-making, because it has not involved the necessary separation of criterion weighting from option performance rating.

Conclusion: Person-centered care will advance only when it is accepted that the central constructs are formative indexes and measurement of them needs to be individually preference-sensitive, not just group preference-based and therefore at least dually-personalised.