The well-being valuation model (WBVM) is built upon several quantitative methods and is at the foundation a quasi-experimental method focused on computing the difference- between expected and actual trends of the study population. The relevant trends in this method apply to well-being risk factors, disease incidence, and disease prevalence. The trend specific differences are interpreted as the well-being improvement program’s effect and are monetized utilizing the total claim costs of the study population. The following subsections describe the way in which each quantitative method fits within the WBVM.
The challenge in accurately valuing the entirety of all effects from risk change is indirect relationships risks demonstrate with one another. Elements of the input–output (IO) model methodology were integrated into the WBVM due to the mathematics and underlying theory supporting quantification of the full complement of interactive relationships or interdependencies between well-being risks, chronic conditions, and age (collectively referred to as inputs). The mathematical structure of the IO model explicitly accounts for the interdependencies, or correlations, between modeled factors and then non-parametrically quantifies the interaction across all inputs (Raa
2009). The resulting values, which are termed
final demand estimates, reflect a set of input-specific estimates quantifying the one-time or sustained increase in population level risk attributed to not modifying prior risk levels.
The input–output model is typically used in regional economic analyses to model the impact of a change in the supply of one input on the output quantity of a good, and vice versa, and by extension, the impact to all inputs and goods within the regional economy. The impact results, or multipliers, are then often used to estimate the indirect benefits or costs of changes to the input and or output structure of the regional economy (Santos et al.
2013; Wiedmann
2009; Santos and Haimes
2004; Ritchie and Dowlatabadi
2014; Jewczak and Suchecka
2014). A recent study demonstrated an approach in which an input–output model was utilized to estimate the monetary impact to a region due to employee absenteeism at a local firm (Bankert et al.
2015). Other than this study and the current application, we are not aware of any other research that has integrated input–output modeling within well-being research to capture the interactive and marginal indirect effects of multiple risks over time on healthcare costs.
To calibrate application of the IO open model for well-being valuation, Spearman’s rho was calculated between the baseline and report year distributions of a given two-factor set. By using Spearman’s rho as the measure of correlation between factors over time we assumed a monotonic relationship between factor distributions (Hauke and Kossowski
2011; Spearman
1904). Calculated rho values were assumed to represent the fundamental physiological relationship between the factors and accordingly, be stable across time, customers and risk profiles. As an example of this methodological component, the Spearman’s rho between body mass index and life satisfaction level was 0.21; all other input to input correlations were measured in this manner. The Spearman’s rho values were aggregated into a 53 by 53 matrix and inverted in the statistical software R.
Specifying the correlation matrix involved quantifying the year-over-year trend in the average prevalence of each input. The input specific trend rates were used to form the vector of expected risks. The vector was comprised of the prevalence estimate for each dichotomous risk input and mean for all other inputs. For training, two versions of the vector were created; one based on the actual year-specific well-being risk distributions and the other based on expected risk trends. In terms of the latter vector, a zero inflated Poisson (ZIP) model was used to estimate the expected number of risks based on individual level baseline characteristics, including demographic, chronic condition and well-being risks as the covariates and actual follow-up values as the response. A trend was then computed for each well-being risk included in the correlation matrix as the average of individual-level risk trend values, which in turn were computed as the ZIP estimated value divided by the actual baseline value. Last, the derived trend estimate was multiplied by the actual average baseline risk value of the study population to form the expected follow-up risk level. All of these expected risk values constituted the final demand vector that was multiplied through the correlation matrix.
The multipliers were computed as the ratio of final demand estimates to the observed value of each input. Each multiplier quantified the percent contribution to the input trend due to the interaction among all of the modeled inputs. In other words, the multipliers represented the expected (1 year) trend in input risk (a) when the risk was left unchanged and (b) after accounting for the influence of all other modeled inputs and marginal indirect effects—externalities—therein. Alternatively, the multipliers can be interpreted as a ‘preventative’ effect since a preventative procedure, such as the flu shot, has both direct and indirect benefits. In the case of a flu shot, the direct benefit is a significantly reduced likelihood of contracting the flu; the indirect benefits include lower likelihood of developing other illnesses due to compromised immune system, higher quality of life during the flu, and decreased time out of work.
3.2 Sources of value and effects for monetization
The monetary valuation of well-being change in terms of health care cost was determined from previously published literature and expected less actual costs from medical and pharmacy claims data (2010 and 2011). The expected costs were expressed in 2010 US dollars having utilized the medical Consumer Price Index from the Bureau of Labor Statistics. The effects of well-being change within WBVM are comprised of four sources: reduction in health care spend among the non-diseased, reduction in the likelihood of developing chronic disease, reduction in medical spend among newly diseased members and last, reduction in spend among those with disease.
3.2.1 Reduction in health care spend among the non-diseased
The marginal cost of being at risk for one or more of the well-being risks within a risk category was the basis for estimating savings due to the reduction of risks among the non-diseased. Studies have shown that people with well-being risks such as physical inactivity, smoking and obesity have higher utilization in terms of hospital stays, physician visits and prescription drugs (Pratt et al.
2000; Raebel et al.
2004; Harrison et al.
2012; Shi et al.
2012; White et al.
2013). Given the existence of these interactions and their manifestation in increased health care utilization, we chose to value reduction in the number and composition of well-being risks based on the differentially lower level of observed health care cost among those with no such risks or disease.
Two steps were followed to estimate the value of reduced risk, or alternatively, the additional cost associated with having one or more risks. In the first step, a ratio was computed of the PPPY medical expenditure for members with at least one risk within a risk category to the PPPY for members with no risks and no disease during the same time period. This ratio, or relative percent difference in cost between the two states (with risk/without risk), for each of the evaluated eight well-being risk factors is listed in Table
2.
Table 2
Relative increase in cost and expected growth rates of well-being risk factors
Relative cost (%)a
| 44 | 10 | 16 | 20 | 11 | 55 | 4 | 15 |
Expected multiplierb
| 0.129 | 0.147 | 0.126 | 0.119 | 0.153 | 0.151 | 0.132 | 0.151 |
The second step of the process for valuing risk reduction among the non-diseased was to monetize value in reducing the number of current risks and preventing development of additional risks. To estimate the number of risks reduced and avoided, there was a need to create an expected level of risks in the next measurement period that could be compared to the realized level of risks. The trend process described earlier concerning multipliers were applied to determine the difference between expected and actual number of risks. This difference was then multiplied by the marginal cost of having at least one risk in the category, aggregated over all members having the risk and then summed over all evaluated risk factors to derive cumulative savings within this component. Since members could have more than one category at risk, we specified a weighted average summation of risk category savings.
3.2.2 Reduction in disease incidence rate
The valuation of this savings component is based on evidence that a direct relationship exists between the composition and number of well-being risks, or more generally, well-being risk profile, and the probability of developing disease, and further, having disease increases the likelihood of developing co-occurring disease ( Wang et al.
2006). People that have chronic conditions spend significantly more in total health care expenditures than those without such conditions. In fact, it has been found that medical costs are approximately double for those with one chronic disease, compared to non-diseased persons, and costs increase exponentially as co-occurring diseases develop (Stanton
2006). Therefore, this method also incorporated savings attributed to a lower likelihood of chronic disease development and subsequent health care costs due to improved well-being risk profile.
Calculating the prevented risks required an estimate of the number of risks that would have been expected to occur in the next year absent intervention. This was achieved utilizing the same methodology detailed in Sect.
3.1. The expected prevalence by risk was then multiplied by risk-specific coefficients derived from logistic models estimating the impact of the expected change in risks on the likelihood of becoming newly diseased. Separate regression models were used to estimate the probability of developing each chronic disease including diabetes, coronary artery disease, chronic obstructive pulmonary disease, heart failure, and asthma. Independent variables in each model included demographic variables age and gender, chronic diseases excluding the dependent variable, and well-being risks aggregated to the factor level (see Table
3). Estimated coefficients from the models estimated on the training population were used to estimate change in the population’s likelihood of developing chronic disease based on changes in aggregate well-being risks by category, holding demographics and disease prevalence constant at their levels.
Table 3
Logistic model coefficients for the model estimating diabetes incidence
Intercept | −11.51 | 19.155 | <0.0001* | |
Current health risk | 1.28 | 9.405 | 0.002* | 3.61 |
COPD (1 = yes) | 0.358 | 0.222 | 0.6369 | 1.43 |
Gender (1 = female) | −0.0002 | 0.0 | 0.9994 | 1.00 |
Age, natural log
| 1.40 | 4.270 | 0.039* | 4.05 |
Asthma (1 = yes) | 0.227 | 0.285 | 0.594 | 2.89 |
Health maintenance risk | 0.270 | 0.515 | 0.473 | 1.31 |
Support risk | 0.771 | 6.188 | 0.013* | 2.16 |
An estimate of disease incidence avoided due to the reduction in the population level well-being risk profile was calculated as the product of the change in probability and number of non-diseased members within the population. A monetary value was then derived by multiplying the estimate of incidences avoided by the net baseline PPPY cost associated with having the disease; the net computation involved deducting the average annual cost of a diseased person from the average annual cost of a non-diseased person.
3.2.3 Reduction in cost of newly diseased
Although reduction in well-being risks reduces the likelihood of developing disease, all incidences will not be avoided. There is value, though, in terms of lowering well-being risk among those members who become newly diseased. This value was captured as the difference in health care cost among these members by differential risk severity level. Specifically, the percent cost difference among newly diseased members by disease and presence versus absence of a risk in a given factor was calculated. Savings were then computed as this difference multiplied by the average diseased PPPY and aggregated across all newly diseased members.
3.2.4 Reduction in cost among the diseased
The WBVM quantifies savings attributed to well-being risk reduction among diseased members using a similar methodology described in the previous section. Due to the low number of members eligible for this component of the savings analysis, a published estimate from the literature that utilized a matching-based methodology was chosen (Wells et al.
2012). The savings estimate from this study of US$24.67 per diseased member per month or US$294.07 PPPY (4.73 % savings rate) was applied to all actively engaged diseased members.
3.3 Coarsened exact matching
A component of WBVM that is used in the instance where administrative claims are not available for a study population is Coarsened Exact Matching (CEM). CEM is a bias-reducing, quasi-experimental matching method employed to quantify a causal effect between cohorts (Iacus et al.
2009,
2011; Sidney et al.
2015). Studies have found CEM to yield causal estimates of a treatment with lower variance and bias of differing sample sizes relative to PSM (Wells et al.
2012; King et al.
2011). CEM was not used in the WBVM presented here but is an available component for the method in order to match the training and study populations to derive weights applicable to the study population.