• Center on Health Equity & Access
  • Clinical
  • Health Care Cost
  • Health Care Delivery
  • Insurance
  • Policy
  • Technology
  • Value-Based Care

In Light of Wellness Program Findings, Employees Should Be Able to Opt Out of Wellness Screenings

Article

Since wellness programs have not demonstrated any meaningful cost savings or outcomes improvement, the Affordable Care Act policy allowing financial coercion by employers to force employees into unlicensed, unregulated, wellness programs should be reconsidered. And, yet, a bill currently awaiting a floor vote in Congress would allow greater financial coercion by employers.

Prior to 2010, it had been widely assumed that the health and financial benefits of workplace wellness (risk assessments, biometric screening, and coaching) were so substantial that the Affordable Care Act (ACA) provision allowing employers to tie large proportions of insurance coverage to wellness program participation, or to measured improvements in health outcomes, was not even debated before being passed into law. It was “the one part of ACA that everyone agreed on” [including this author], as Senator Lamar Alexander, R-Tennessee, quite accurately stated.

This consensus assumption was based on 2 pieces of evidence. The first was Safeway’s claim that wellness reduced its health spending 40%, a claim that garnered so much publicity that the wellness provision of ACA became known as the “Safeway Amendment.” It was later revealed that Safeway didn’t have a wellness program during the period in which the savings from it were claimed.

The second was a meta-analysis showing that programs at large companies had averaged more than $3 in savings for every $1 invested, a finding later invalidated by one of the principal investigator’s own subordinates.

Subsequent to ACA enactment, small-scale findings by researchers and practitioners outside the industry have universally found financial benefits or health improvements to be trivial at best. This plethora of “micro” case evidence, combined with the unpopularity, intrusiveness, equity concerns, and (in some cases) documented harms of workplace wellness, calls for a “macro” review to determine whether wellness has indeed saved money for the United States through reducing hospital admissions attributable to events such as heart attacks, diabetes, stroke, and hypertensive events—wellness-sensitive medical admissions, or WSMAs, that the industry claims to address—that should theoretically be avoidable through these programs.

Hypothesis

If, indeed, wellness reduces hospital admissions, Healthcare Cost and Utilization Project (HCUP) data over a sufficiently long period should reveal gradually increasing separation—mirroring the dramatically increasing spending on wellness over that period—between the rate of WSMAs in the exposed population versus the nonexposed population of similar age, which was not insured privately.

This difference-of-differences should, by the end of the study period, be substantial enough to either show savings or at least be “on track,” via an increasing difference-of-differences in admissions to project a future breakeven.

Methods

This easily replicable and totally transparent analysis was conducted, as follows, in 2 parts. Both parts required accessing the HCUP database from 2001 to 2014, which tallies nearly all hospital admissions in those years and is designed to facilitate analyses such as this.

The first part of the analysis was to discern separation between the study population and the control population.

To do that, overall (nonmaternity) admissions were split into 3 categories. The first was what HCUP calls “privately insured” (meaning mostly <65 years old employees or dependents, but with individually insured admissions, too). The second group consisted of admissions paid for by Medicaid. The third group was people who were uninsured or self-paying. In total, those groups should capture virtually the entire nonexposed <65 years old population.

These control populations are notable for lacking consistent access to preventive care and having generally worse socioeconomic status than the insured population. Therefore, their WSMAs should underperform the privately insured population. This prima facie confounding bias means that, had there been significant separation, attribution to wellness (versus social determinants of health) could be arguable.

Next, the International Classification of Diseases, Ninth Revision, codes for WSMAs were separately accessed. These codes were (inclusive of all 5-digit codes) 250xx for diabetes, 410xx for acute myocardial infarction, 434.91 for stroke, and 401xx for hypertensive events. Tallying WSMAs is a widely accepted approach in wellness. The past chair of the Health Enhancement Research Organization (HERO, the wellness industry promotional association), Ron Goetzel, PhD, agrees: “The idea of analyzing claims for conditions likely to be most readily influenced by health promotion programs is sensible.” The official HERO Program Measurement and Evaluation Guide recommends it.

The WSMAs were split into the same 3 categories. The WSMAs were then divided by the number of total admissions for all categories. Four specific reasons dictated why total admissions would make a better denominator than a simple rate per 1000 population. First, during these 14 years many factors affected all admissions—better therapies, a move to outpatient care, increased adoption of high-deductible health plans (HDHPs), re-coding from admissions to observation days, and increased cost of hospitalizations would be among them.

Second, if there is a systematic error in the HCUP data collection, it would presumably affect all admissions equally and therefore this methodology would moot it. For instance, between 2005 and 2006 the population rates all inflected, for unknown reasons, but the inflection does not affect this methodology because it took place in all 3 populations.

Third, the populations themselves are not static, as people change payers. For example, people who were previously uninsurable could, following ACA passage, sign up for coverage either through Medicaid or the exchange. If those people are, plausibly, sicker than average, they will have more all-cause admissions than average. Making total admissions the denominator is, in some sense, a risk adjustment. Of course, the risk adjustment could go the other way too, as some of the previously uninsured had chosen not to buy insurance. They would have both low all-cause admissions and low WSMAs. Likewise, people in HDHPs could be expected to have lower rates of all admissions. Since the numerator and the denominator move together, contamination of the cohorts is therefore an unlikely confounder.

The actual data is appended to Figure 1 (see Results section), and the spreadsheets themselves are available upon request.

The second part was a “number needed to decrease” analysis. The difference-of-differences—the change in the number of admissions over the period versus the control&mdash;was compared with the change in spending on wellness programs over that period.

Results

Over this 14-year period, in which the wellness industry grew from being too small to merit a published estimate of its size to $7.8 billion a year, there was no corresponding reduction in WSMAs in the increasingly exposed population of privately insured people on an absolute basis (2014 vs 2001) or compared with the benchmark of total nonexposed populations under age 65, as Figure 1 shows. (Medicare, though not useful as a control due to age, trends much more positively than either despite having zero access to workplace wellness.)

Those 3 control populations—the uninsured, Medicaid, and self-pay/other&mdash;could logically be expected to do worse, owing to, presumably, worse social determinants of health. Yet the year-over-year and endpoint-to-endpoint difference is not great enough to support any hypothesis other than that no input or exogenous shock or demographic change over that period had any substantial differential impact on trend.

Figure 1. Privately insured WSMAs as a % of all admissions vs Medicaid/uninsured/self-pay vs Medicare

The second part was a “number needed to decrease” analysis. The difference-of-differences was compared with the change in spending on wellness programs over that period.

The number of relevant admissions in 2014, according to HCUP, was 398,790. HERO estimates that WSMAs cost $22,500, for a total of $9 billion in spending. By contrast, the workplace wellness industry grew to $7.8 billion by 2014. No one knows what the industry size was in 2001 because it wasn’t tracked. As a plug figure (and changing this assumption would not materially affect the conclusion), assume $1 billion. For employers to break even on wellness through savings on admissions (and before considering the list of healthcare costs that HERO acknowledges could increase as a result of wellness), the 2014 admissions figure at $22,500 per admission would have to have fallen by more than 300,000, creating a trendline that would approximate that shown by the black dotted line in Figure 2.

Figure 2. Approximate WSMA decline required to break even on wellness spending

Discussion and Limitations

This study is intended only to show a major magnitudinal and directional result, not a precise impact of wellness. Hence limitations affecting precision would not affect the magnitude or direction of the result.

One limitation is that the percentage of Americans older than age 65 who are employed full- or part-time has increased from 13% to 19% over the period in question. Some number of them would still be “privately insured.” However, because the Medicare WSMA trend is so positive, the bleed of seniors (in the ratio of WSMAs to all their admissions) into the privately insured subset should make the latter trend more favorable.

A third limitation is that while virtually all people subject to wellness would be in the privately insured cohort so that sensitivity is close to 100%, specificity can’t be estimated because this industry’s opacity makes exact wellness program enrollment figures difficult to discern—and wellness programs vary from employer to employer and by employer size. One consequence of this limitation is that the lower the figure one assumes for specificity (penetration of wellness programs), the greater the excess of spending on wellness programs over spending on WSMAs in the subset of the population subject to these programs, to the point where even elimination of all WSMAs in the penetrated population could not result in a breakeven.

For example, if half the privately insured population is exposed to wellness, the $7.8 billion in spending would substantially exceed the $4.5 billion in WSMAs.

This is consistent with practice: as noted in the discussion below, a typical spending level of $100 to $150 per employee per year would far exceed savings even in award-winning programs.

It is important to recognize that [wellness] should increase the use of certain services, such as preventive and screening services, certain chronic medications, and outpatient visits. It is even possible to see a rise in ER and urgent care visits as well-informed patients learn to get urgent medical care.

A fourth limitation is that other medical expenditures could decrease. However, there is no evidence of that. Typically, programs require more checkups and other preventive care, and costs increase. HERO wrote:

A final limitation is that wellness may “pay for itself” in some other way, such as employees having higher productivity or fewer absences. That is beyond the scope of this analysis, but wellness programs aimed squarely at cardiometabolic risk reduction should see some impact on relevant WSMAs.

Review of Related Literature

This “macro” finding meshes with the case study literature published within the last 5 years, results self-reported by HERO and leading vendors, and wellness industry promoters’ own comments.

Four of those case studies were observational. An analysis of the Barnes Hospital wellness program revealed a reduction in hospitalizations but no savings. Further, the baseline rate of WSMAs (using approximately the same definition used in the current analysis) was roughly 24 per 1000 employees per year or 27% of all admissions—both figures being many times the national average. This makes it possible to attribute some or all of the reduction in hospitalizations to regression to the mean. An analysis of the Connecticut state employee program revealed an actual claims cost increase, which the authors attributed to more annual checkups and other preventive care, included required mammograms for women under 40. An analysis of 373,578 medical records correlated lower risk factors with higher spending.

Many studies, including the widely cited meta-analysis by Baicker et al estimating a 3.27-to-1 ROI, “show savings” by comparing participants to non-participants. However, a recent study by Jones et al, controlling for the impact of self-selection by would-be participants, found no impact on health behaviors or outcomes in the first year.1 It concluded that the participant-vs-non-participant study design was itself invalid.

Jones’ conclusion regarding the invalidity of that study design is consistent with observational findings: In the 3 instances in which it was possible to compare participants vs nonparticipants study designs with a known benchmark, the entire separation could be attributed to the study design. For example, in 1 case, would-be participants showed roughly a 20% favorable separation in costs versus nonparticipants after 2 years, even though there was no program to participate in during that period. All 3 studies were conducted by wellness industry promoters, ruling out investigator bias.

Jones’ finding is also consistent with the wellness industry trade publication’s acknowledgement that “randomized control trials show negative ROIs [return on investments].” The only randomized control trial published subsequent to that meta-analysis also found little to no impact of incentivized weight loss, implying no savings and a negative ROI.

The principal finding of this study is also broadly consistent with the HERO’s own finding that wellness avoided less than $12 per employee per year in reduced hospitalizations, before program fees (typically 5 to 12 times that figure) and other costs are considered.

The finding is consistent with the results achieved by the most recent winners of the wellness industry’s award for program achievement. The 2015 winner was not able to improve biometric status.2 Meanwhile, the 2016 winner’s program acknowledged a deterioration in the health status of the participants, mooting any claims of savings.

The finding is also consistent with the Kaiser Family Foundation’s review of health risk assessments, where it was found that company-wide completion rates for this tool, considered a pillar of wellness, correlated with higher health spending, and at the individual level, completion also correlates with higher spending.

The finding is also consistent with commentary in the lay media, which is uniformly negative, regardless of the political leaning of the publication.

The finding is also remarkably consistent with the admissions of HERO’s own principals, such as “Only … 100 programs [have succeeded while] thousands … are not getting good outcomes,” and that even those programs “done right” achieve only a “1% to 2% improvement,” after “2 to 3 years.”

Conclusion and Policy Implications

Because wellness has not demonstrated any meaningful cost savings or outcomes improvement and even award-winning programs are associated with deteriorations in health, the ACA policy allowing financial coercion by employers to force employees into unlicensed, unregulated, wellness programs should be reconsidered.

And, yet, a bill currently awaiting a floor vote in Congress, the Preserving Employee Wellness Programs Act, would allow greater financial coercion by employers, overriding the prohibition against involuntary medical exams and inquiries in the Americans with Disabilities Act. It would also allow employer programs to access the DNA of employees and family members. Even evidence from DNA testing vendors does not support the value of DNA testing.

Quite the opposite: employees should be allowed to opt out of these “voluntary” wellness screening programs with no financial forfeiture. Screenings should also be scheduled, and blood panels chosen, according to age-adjusted clinical guidelines, instead of requiring often-inappropriate screening blood values at inappropriate annual intervals for everyone, which increases the likelihood of expensive and hazardous overdiagnosis and overtreatment—and does not reduce relevant hospitalizations by a meaningful amount.

Notes

1. The principal investigator of that study, Damon Jones, PhD, is an associate professor at Harris School of Public Health, where Katherine Baicker, PhD, is a dean—a relationship likely unique in the annals of health services research that allows investigator bias to be ruled out.

2. See Figure 4 on page 9 of The McKesson Koop Award 2015 C. Everett Koop Award application, listing the “before” and “after” results of biometric screenings. The deteriorations in 3 of the biometric indicators almost exactly offset the improvements in the other 2 even without considering that dropouts lost to follow-up might tip the scales in the negative direction.

Related Videos
Mei Wei, MD.
Screenshot of Susan Wescott, RPh, MBA
Glenn Balasky, executive director of the Rocky Mountain Cancer Center.
4 KOLs are featured in this series
4 KOLs are featured in this series
Screenshot of an interview with Stuart Staggs
Related Content
© 2024 MJH Life Sciences
AJMC®
All rights reserved.