Defining recovery as a dichotomous variable enables analysis at the group and individual level, with some advantages. In dichotomous recovery, a patient is recovered if their postoperative score equates to or exceeds a threshold value (usually their pre-surgical value), either overall or in each domain. Recovery can be compared between groups (by comparison of the prevalence, and hence odds ratio, of recovery) or at the individual level. Importantly, it can assess recovery within differing domains. Unlike composite scores, which collapse all domains into one overall score, dichotomous scores have the ability to ‘drill down’ and identify in which (an how many) domains a patient failed to recover. It preserves the ability to assess recovery at four levels; overall recovery, within each domain to see where recovery has failed, within each subdomain to identify the particular problem (e.g. is the failure of functional recovery due to eating, dressing, standing or walking), and the severity of the problem indicated by the scores, or by the number of tests within each subdomain that fail to recover (Fig. 5.2), all of which are clinically relevant. This has implications in both the clinical and research setting, as it enables targeted clinical intervention or assessment of impact of intervention respectively.

Dichotomisation of a continuous variable must always be substantiated and none more so than with recovery. Recovery may be seen, as with many physiological variables, to be a continuum (i.e. 50 % recovered, 75 % recovered etc.). Thus, to maintain statistical rigor and minimise bias, it is argued that its data collection and statistical assessment should follow suit. In research, recovery data is often collected as a continuous variable which preserves data integrity and maximises internal validity and power with the lowest sample size [71]. Comparison of recovery between groups is by comparing the difference in mean change score between intervention and control groups in order to audit or research an intervention. However, the utility of recovery as a continuous variable for clinical decision-making is limited. Whilst recovery may be a continuum, the outcome of a clinical decision is binary – the patient has or has not recovered enough to finish the crossword or safely return to work (patient focus), or to be discharged (institution focus). A difference in mean change scores (derived from analysing continuous variables) does not directly indicate a tangible clinical effect, whilst recovery as a dichotomised outcome has direct clinical utility. When comparing recovery between two groups as a continuous variable, a difference in mean change scores may indicate a statistical difference in recovery between the two groups, but does not indicate whether this has clinical implications. By comparison, when comparing recovery between two groups as a dichotomous outcome, a difference in incidence or odds ratio of recovery has direct clinical consequences in terms of resource utilisation. In addition, the issue regarding the increased sample size resulting from dichotomisation is irrelevant in the clinical setting where the assessment is on an individual patient. This reinforces a limitation of translating research tools, with their definition of recovery as a continuous variable, into the clinical setting.

It is argued that by dichotomising a continuous variable, data richness is lost i.e. dichotomisation does not indicate the magnitude by which a patient may fail to recover but rather just that the postoperative threshold value for recovery has not been achieved. However, this limitation can be mitigated by the collection of raw data in its continuous form (where applicable) and subsequent secondary dichotomization for analysis. This enables analysis of a clinically relevant outcome (i.e. has this patient recovered or not), with secondary analysis of the raw data to determine the magnitude (i.e. by how much has the patient failed to recover?).