WMHP users were coded as such in this study based upon their use of at least 1 WMHP service. WMHP providers submitted diagnostic impressions that were then mapped to ICD-10 diagnosis codes by Lyra Health clinicians. A detailed mapping from diagnostic impression to ICD-10 diagnosis codes is included in the Appendix Table A1 .
To apply the Aon cost efficiency measurement process, WMHP users were matched with a comparison group of nonusers, composed of eligible members at the same employers with closely matched geography, demographics, and medical and mental health comorbidities, for the same time periods. A derivation of coarsened exact matching32 was used to match cohorts. As originally described by Iacus et al, coarsened exact matching first involves dividing members into meaningful categories selected for each matching factor of interest.
Specifically, members were first divided into age groups, gender categories, geographic area categories, and according to the existence or not of several medical conditions, as noted below. Then, all members of the treatment and comparison groups who fell into the same categories were retained for the analysis; the rest were excluded. The Iacus et al32 method also suggests using case weights to account for the proportion of treatment and comparison group members who are in each factor category, then using their original data values (not the indicators of which factor categories they fell into) in the subsequent statistical analyses.
The Iacus et al method was simplified for this analysis, by avoiding the use of case weights. Age is the only continuous measure in this data set, so individuals were matched on tightly constructed age groups, which still produced a highly balanced set of WMHP users and comparison group members for analysis. More specifically, individuals were matched by gender first; then they were matched to others within ±3 years of their ages.
Individuals were then matched on presence or absence of 22 diagnosed medical conditions (Tables 1 and 2 ), and combinations of selected conditions. The chronic condition indicators considered for each member were based on primary (first listed) medical diagnostic codes, using the Chronic Condition Indicator and Clinical Classifications Software developed by the Agency for Healthcare Research and Quality Healthcare Cost and Utilization Project.33 When members had more than 1 mental health condition, a hierarchy was applied.
Specifically, when 2 or more disorders were present, members were coded according to the one that is typically thought to be more impactful in nature, resulting in greater levels of disability and incurring the most cost. Based upon the existing literature,2 (link),4 (link),5 (link) mood disorders were anticipated to incur more costs than anxiety and adjustment disorders, which were in turn anticipated to incur more costs than attention-deficit disorders.
Next, WMHP users were matched to nonusers in the same geographic areas, when possible. First, an attempt was made to match members residing in the same metropolitan statistical area (MSA), based on a list of over 200 such areas across the United States. When a within-MSA match could not be found, members were matched at the state level. When that was not possible, members were not matched geographically, but were matched on the other factors mentioned above, searching across the country for the best demographic and condition level matches.
When no matches could be found (typically when members had a rare combination of disease and location values), individuals were removed from further analysis. For WMHP users who could have matched to more than 1 comparison group member, only 1 matching comparison group member was randomly selected for inclusion into the analyses. SeeFigure 1 for other inclusion and exclusion criteria and associated sample size reductions.
To apply the Aon cost efficiency measurement process, WMHP users were matched with a comparison group of nonusers, composed of eligible members at the same employers with closely matched geography, demographics, and medical and mental health comorbidities, for the same time periods. A derivation of coarsened exact matching32 was used to match cohorts. As originally described by Iacus et al, coarsened exact matching first involves dividing members into meaningful categories selected for each matching factor of interest.
Specifically, members were first divided into age groups, gender categories, geographic area categories, and according to the existence or not of several medical conditions, as noted below. Then, all members of the treatment and comparison groups who fell into the same categories were retained for the analysis; the rest were excluded. The Iacus et al32 method also suggests using case weights to account for the proportion of treatment and comparison group members who are in each factor category, then using their original data values (not the indicators of which factor categories they fell into) in the subsequent statistical analyses.
The Iacus et al method was simplified for this analysis, by avoiding the use of case weights. Age is the only continuous measure in this data set, so individuals were matched on tightly constructed age groups, which still produced a highly balanced set of WMHP users and comparison group members for analysis. More specifically, individuals were matched by gender first; then they were matched to others within ±3 years of their ages.
Individuals were then matched on presence or absence of 22 diagnosed medical conditions (
Specifically, when 2 or more disorders were present, members were coded according to the one that is typically thought to be more impactful in nature, resulting in greater levels of disability and incurring the most cost. Based upon the existing literature,2 (link),4 (link),5 (link) mood disorders were anticipated to incur more costs than anxiety and adjustment disorders, which were in turn anticipated to incur more costs than attention-deficit disorders.
Next, WMHP users were matched to nonusers in the same geographic areas, when possible. First, an attempt was made to match members residing in the same metropolitan statistical area (MSA), based on a list of over 200 such areas across the United States. When a within-MSA match could not be found, members were matched at the state level. When that was not possible, members were not matched geographically, but were matched on the other factors mentioned above, searching across the country for the best demographic and condition level matches.
When no matches could be found (typically when members had a rare combination of disease and location values), individuals were removed from further analysis. For WMHP users who could have matched to more than 1 comparison group member, only 1 matching comparison group member was randomly selected for inclusion into the analyses. See
Free full text: Click here
