From simple to even simpler, but not too simple: a head-to-head comparison of the Better-Worse and Drop-Down methods for measuring patient health status

Abstract

Background: We recently developed a novel, preference-based method (Better-Worse, BW) for measuring health status, expressed as a single metric value. We have since expanded it by developing the Drop-Down (DD) method. This article presents a head-to-head comparison of these two methods. We explored user feasibility, interpretability and statistics of the estimated coefficients, and distribution of the computed health-state values.

Methods: We conducted a cross-sectional online survey among patients with various diseases in the USA. The BW and DD methods were applied in the two arms of the study, albeit in reverse order. In both arms, patients first performed a descriptive task (Task 1) to rate their own health status according to the 12 items (each with 4 levels) in the CS-Base health-outcome instrument. They then performed Task 2, in which they expressed preferences for health states by the two methods. We then estimated coefficients for all levels of each item using logistic regression and used these to compute values for health states.

Results: Our total sample comprised 1,972 patients. Completion time was < 2 min for both methods. Both methods were scored as easy to perform. All DD coefficients were highly significant from the reference level (P < 0.001). For BW, however, only the second-level coefficient of "Cognition" was significantly different (P = 0.026). All DD coefficients were more precise with narrower confidence intervals than those of the BW method.

Conclusions: Both the BW and DD are novel methods that are easy to apply. The DD method outperformed the BW method in terms of the precision of produced coefficients. Due to its task, it is free from a specific distorting factor that was observed for the BW method.

Keywords: Health status; Health-related quality of life (HRQoL); Measurement model; Patient-reported outcome measures (PROMs); Preference-based; Values.

Fig. 1

The 12 items of the generic health-outcome instrument CS-Base, each with four levels, as depicted in the HealthSnApp (an application for mobile phones)

Fig. 2

Screenshots of the CS-Base from the HealthSnApp during Task 1. In this descriptive task, all health items were listed in interactive (rotating) boxes presented on a single screen. When a patient selected the interactive box for a specific item, the box displayed response options. For example, when a patient selected the “Fatigue” box, the display shifted to offer the following response options: “Not tired,” “A little tired,” “Quite tired,” and “Very tired”

Fig. 3

Screenshots of Task 2 from the Better-Worse (BW) and Drop-Down (DD) assessment and judgment tasks. For the BW method, respondents (i.e., patients) compared their own health states to five slightly different, alternative health states. With the exception of only two items, the alternative health states portrayed in Task 2 of the BW method did not differ from the actual health states as reported by the patients in Task 1. One of these items depicted an improvement of one level relative to the patient’s actual health state (depicted as a green box). Another item showed a reduction of one level relative to the patient’s actual health state (depicted as a red box). For the DD method, patients made multiple selections (2–5 times) of items at the levels that hindered or disturbed them the most. They did this by swiping (dropping down) the level and moving the item one level lower (i.e., better)

Fig. 4

Schemes representing the steps taken to generate data for the Drop-Down analysis. The example reflects rated levels for the 12 items (with 7 levels being higher than the reference Level 1) of the CS-Base ePROM (left). Responses were stored on the server after five drop-downs (middle). The final data for the analysis were obtained after inserting postulated health states (right)

Fig. 5

Distribution of coefficients and their 95% confidence intervals for the Better-Worse (BW) and Drop-Down (DD) methods

Fig. 6

Distribution of computed values, based on the estimated regression coefficients, of all health states reported by the patients for the Better-Worse (BW: 1,184) and Drop-Down (DD: 1,123) methods

Fig. 7

Computed values for health states in the CS-Base ePROM, based on the estimated regression coefficients from the Better-Worse (BW) and Drop-Down (DD) methods for all possible health states (16,777,216; depicted as light blue dots) and the common BW and DD states (328) reported by the patients in this study (dark blue dots)