Rand Report on Workplace Wellness: What Employers Must Know (Part II)

Q&A about the Rand Report on Workplace Wellness (Part II)

What Employers Must Know

Read Q&A about the Rand Report on Workplace Wellness (Part I)

Questions and Answers: A Conversation with Soeren Mattke and Hangsheng Liu

How were the final sample sizes (N) determined?

Due to the potential for future publication of the study analysis, we cannot provide additional information that was not included in the final report. However, we can point you to Table B1 (p. 122) in the appendix and to the footnotes throughout the report for sample size data.

In Table B3 (p. 126), we see a column labeled, “Variables Not Balanced After Matching,” but no detailed percentages or averages are stated. Can we see the statistics on the matching variables before and after the match?

None were imbalanced between groups.

Other risk factors (e.g., glucose, blood pressure, alcohol consumption, overall health status, fruit/vegetable consumption) were provided in the data but not included in the report. Why?

There were several reasons. First, due to the length of the report we were unable to include all risk factors. We prioritized the list and analyzed the most common risk factors. Data availability was also a factor.

In the obesity analysis, Figure 4.17 (p. 50) indicates a 14% difference between the participant and nonparticipant rates. How does this relate to a 1-pound reduction?

Using the same regression analysis and underlying sample, we simulated a final analytic sample supposing all participants participated in the program all five years. This simulated hypothetical cohort of participants and nonparticipants was then projected over time. The dichotomy between the actual regression and the simulation makes it difficult to explain how the numbers were derived but provides an elegant way of translating parameter estimates in a nontechnical way. Since the model runs on BMI units, we then assumed the characteristics of a standard man and woman (based on CDC data) and translated the change in BMI to change in pounds. This finding of a 1 pound per year reduction is significant since it represents the average weight loss of participants compared to nonparticipants on a population level (i.e., 1 lb. times the number of participants). Furthermore, participants continued to lose an average of one pound/year in the first and second years after the year of participation.

Regarding the cost analysis (pp. 53-57). a. Is this cohort over the whole time horizon? b. Were outliers excluded? 3. Did you use company cost or company + employee cost?

a. Not all employees were involved for all years.

b. We used 99% threshold for outliers. 3. Analysis was based on allowable claims data, or company cost + employee cost.

Can you explain further the conclusion that incentives were associated with significant improvements in smoking, BMI and exercise, yet the effect size was small (p. 87)? How were these conclusions determined?

The challenge with this estimation was that the employers included in the analysis had little variation with incentives offered, both across employers and within each employer over time. Therefore, we had a large sample size with little variation to run the regression. Figure 5.19 (p. 87) indicates that higher incentives impact reduction in BMI with significant effect. When translated to pounds, we still detect a significant effect. However, given the small effect size and little variation, it is difficult to make a strong statement on the effect of incentives on health outcomes. Furthermore, employers in the sample used incentives for participation only, not outcomes.

Did you adjust for the differences in benefit plans between employers?

No, we did not have access to benefit information, but we did match by employer, candidate year, employee comorbidities, prior utilization and cost. While this limitation could confound the results, the problem should be minimal. One might speculate that analysis of the benefit design might show that the impact was underestimated. This would make and interesting empirical study for future research.

Was there any adjustment for cost differences for the same service in different facilities and regions of the country? For example, an individual living in a high cost region of the country pays $4,000/year with a 6%/year increase, while another living in a low cost region pays half the amount. Even in a difference-in-difference model, the analysis would indicate a cost increase when the % change favors the participant.

Yes, but the analysis matched on baseline cost also. However, this introduces another issue, which is matching on services or on baseline costs, not both. Ideally, the groups would be matched by geographical area, and in one regression analysis geographical region was considered.

How was participation defined? Without a minimal threshold of participation, there may be little to distinguish a participant from a nonparticipant, and thus little expectation of savings.

We used the definition from each respective data contributor, including their definition of minimal threshold, for the comparison group. We did consider the effect that the definition might have on the analysis and assumed program responsibility to engage the participant. With that technique, we arrived at a real-world estimate of the program’s impact. We would like to do further research to include a dose-response curve and examine efficacy of continuous participation.

Was there any attempt to measure presenteeism, absenteeism, or productivity impact other than asking employers what they liked about the program? Do you plan to do so in future research?

We did not have data on presenteeism or absenteeism. However, we would like to have this data for additional analysis. See Section 7.4.2 (p. 109) regarding future research.

Read Q&A about the Rand Report on Workplace Wellness (Part I)


Please leave your thoughts

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: