Intervention analysis and classification: key to health outcomes optimization

Abstract

Purpose

Measures are important to healthcare outcomes. Outcome changes result from deliberate selective intervention introduction on a measure. If measures can be characterized and categorized, then the resulting schema may be generalized and utilized as a framework for uniquely identifying, packaging and comparing different interventions and probing target systems to facilitate selecting the most appropriate intervention for maximum desired outcomes. Measure characterization was accomplished with multi-axial statistical analysis and measure categorization by logical tabulation. The measure of interest is a key provider productivity index: “patient visits per hour,” while the specific intervention is “patient schedule manipulation by overbooking.” The paper aims to discuss these issues.

Design/methodology/approach

For statistical analysis, interrupted time series (ITS), robust-ITS and outlier detection models were applied to an 18-month data set that included patient visits per hour and intervention introduction time. A statistically significant change-point was determined, resulting in pre-intervention, transitional and post-effect segmentation. Linear regression modeling was used to analyze pre-intervention and post-effect mean change while a triangle was used to analyze the transitional state. For categorization, an “intervention moments” table was constructed from the analysis results with: time-to-effect, pre- and post-mean change magnitude and velocity; pre- and post-correlation and variance; and effect decay/doubling time. The table included transitional parameters such as transition velocity and transition footprint visualization represented as a triangle.

Findings

The intervention produced a significant change. The pre-intervention and post-effect means for patient visits per hour were statistically different (0.38, p=0.0001). The pre- and post-variance change (0.23, p=0.01) was statistically significant (variance was higher post-intervention, which was undesirable). Post-intervention correlation was higher (desirable). Decay time for the effect was calculated as 11 months post-effect. Time-to-effect was four months; mean change velocity was +0.094 visits per h/month. A transition triangular footprint was produced, yielding 0.35 visits per hr/month transition velocity. Using these results, the intervention was fully profiled and thereby categorized as an intervention moments table.

Research limitations/implications

One limitation is sample size for this time series, 18 monthly cycles’ analysis. However, interventions on measures in healthcare demand short time cycles (hence necessarily yielding fewer data points) for practicality, meaningfulness and usefulness. Despite this shortcoming, the statistical processes applied such as outliers detection, t-test for mean difference, F-test for variances and modeling, all consider the small sample sizes. Seasonality, which usually affects time series, was not detected and even if present, was also considered by modeling.

Practical implications

Obtaining an intervention profile, made possible by multidimensional analysis, allows interventions to be uniquely classified and categorized, enabling informed, comparative and appropriate selective deployment against health measures, thus potentially contributing to outcomes optimization.

Social implications

The inevitable direction for healthcare is heavy investment in measures outcomes optimization to improve: patient experience; population health; and reduce costs. Interventions are the tools that change outcomes. Creative modeling and applying novel methods for intervention analysis are necessary if healthcare is to achieve this goal. Analytical methods should categorize and rank interventions; probe the measures to improve future selection and adoption; reveal the organic systems’ strengths and shortcomings implementing the interventions for fine-tuning for better performance.

Originality/value

An “intervention moments table” is proposed, created from a multi-axial statistical intervention analysis for organizing, classifying and categorizing interventions. The analysis-set was expanded with additional parameters such as time-to-effect, mean change velocity and effect decay time/doubling time, including transition zone analysis, which produced a unique transitional footprint; and transition velocity. The “intervention moments” should facilitate intervention cross-comparisons, intervention selection and optimal intervention deployment for best outcomes optimization.

Citation

Nkwocha, O. (2019), "Intervention analysis and classification: key to health outcomes optimization", International Journal of Health Care Quality Assurance, Vol. 32 No. 2, pp. 347-359. https://doi.org/10.1108/IJHCQA-03-2018-0066