What Does Lyra’s 3:1 ROI Measure?

Lyra says a Fortune 500 employer saved $3.04 in health-care costs for every $1 invested in Lyra, which it describes publicly as approximately a 3:1 ROI. It also says members who used Lyra incurred $4,138 less in annual health-plan claims than matched members who did not.

The July 2024 single-employer study, commissioned by Lyra, advances the economic analysis in one important respect: unlike the earlier four-year study, it attempts to include the employer's full Lyra cost.

But reconstructing the reported 3.04 requires more information than the report provides.

The $4,138 is straightforward

The study reports annual claims of:

Lyra participants: $5,986
Matched nonparticipants: $10,124
Difference: $4,138 per participant

The $4,138 consists of $2,566 lower mental-health medical spending, $1,056 lower non-mental-health medical spending, and $516 lower prescription-drug spending.

That establishes the numerator used in the economic comparison: matched participants had $4,138 lower annual claims.

The report then states that the ratio of the matched spending difference to “total average fees,” including Lyra costs associated with people outside the matched population, was 3.04.

The 3.04 does not follow directly from the other reported number

Elsewhere, the study says total Lyra session and support fees were $1,162 per participant. The fees include PMPM support costs across eligible members and session costs for Lyra users, including users outside the matched study population.

Using the two disclosed figures produces $4,138 ÷ $1,162 = 3.56, not 3.04.

Conversely, a 3.04 ratio applied to the $4,138 claims difference implies an effective denominator of approximately: $4,138 ÷ 3.04 = $1,361.

That is about $199, or 17%, higher than the separately reported $1,162 per participant.

The report provides a possible explanation but not a complete reconciliation. It says the 3.04 calculation uses all Lyra fees, including costs associated with the non-matched population, and describes 3.04 as the most conservative of several cost-ratio calculations. Its sensitivity analyses produced ratios ranging from approximately 3.0 to 3.7.

What the report does not provide is the calculation that gets from $1,162 to the effective denominator required to produce 3.04.

The 3.04 ROI is not necessarily incorrect, but a purchaser cannot reproduce it from the figures presented in the report.

Which members are in the calculation?

The population descriptions also require careful reading.

The employer had approximately 114,000 members. Before matching, Table 3 shows 4,374 participants and 109,795 nonparticipants. Those figures describe the population before matching; the 109,795 members are not the matched comparison group used to produce the $4,138 result.

After matching, the participant population falls by 17%, to 3,631 members.

The methodology says each participant was matched to the most similar nonparticipant based on demographics, geography, and diagnosed conditions. But the report never separately states the final number of matched nonparticipants used in the cost comparison. It also does not provide enough detail in the summary tables to reconstruct precisely how the total employer fee pool is allocated back to the matched participant population for the 3.04 calculation.

This creates two different population concepts inside the ratio:

The claims difference comes from the matched study population

The cost calculation intentionally incorporates Lyra expenses incurred beyond that matched population.

Including all program costs is directionally appropriate for a purchaser. But it makes the normalization important. A purchaser should be able to see exactly which costs are being spread over which members and why.

The denominator is only one part of the ROI question

Even if the 3.04 arithmetic were fully reconciled, the larger economic question remains in the numerator.

The $4,138 is the entire observed difference between matched users and nonusers. For the ratio to represent savings produced by Lyra, that difference has to approximate what those participants would have cost without Lyra.

The study uses extensive matching. But participation was not randomized, and the report acknowledges that unobserved differences—including socioeconomic factors, job type, reasons for engaging with mental-health care, care-seeking preferences and condition severity—could affect the comparison.

That matters because every dollar of unexplained pre-existing difference that enters the $4,138 numerator also enters the reported economic return.

The condition results make attribution especially important

The study reports lower participant spending in 14 of the 15 clinical-condition categories shown, although not every difference was statistically significant.

The differences include:

• $7,425 lower spending among members with joint disorders;

• $7,837 lower spending among members with lower-back or disc disease;

• $4,367 lower spending among members with cardiovascular disease or hypertension;

• $6,354 lower spending among members with diabetes; and

• $11,806 lower spending among members with neurological disorders.

Mental-health treatment can plausibly influence physical-health utilization through adherence, care seeking, chronic-pain management, substance use, and other pathways. The table therefore does not establish that these differences are unrelated to Lyra.

But neither does diagnosis matching establish that Lyra caused them.

A broad favorable spending pattern across many different clinical conditions makes the counterfactual question more important, not less: were Lyra users lower-cost because of the intervention, or were there remaining differences between the members who elected to use Lyra and those who did not?

That question matters directly to the $4,138 numerator.

Is 3.04 an ROI?

There is also a terminology distinction worth preserving.

The July study describes 3.04 as a cost ratio. Lyra publicly describes the same result as a 3:1 ROI and as $3.04 saved for every $1 invested.

Those formulations are common in health-benefit purchasing, but the calculation is more precisely a gross savings-to-cost ratio: claimed savings divided by program cost.

A conventional net-return calculation would first subtract the investment from the financial benefit. The more important purchaser issue here, however, is not terminology. It is whether the numerator represents savings attributable to the program and whether the denominator can be independently reconstructed.

What should a purchaser be able to reconcile?

Before underwriting a 3:1 return, a purchaser should be able to answer:

• What exact numerator and denominator produce 3.04?

• Why does the study report $1,162 in total Lyra fees per participant when the reported 3.04 ratio implies approximately $1,361 of cost for each $4,138 of claimed savings?

• How many nonparticipants were included in the final matched comparison, and could any control member be used more than once?

• How were PMPM support fees for the approximately 114,000 eligible members and session costs for unmatched users allocated into the final ratio?

• How much of the $4,138 difference represents services shifted into Lyra rather than utilization avoided altogether?

• What evidence supports attributing the broader medical and pharmacy differences to Lyra rather than remaining differences between users and nonusers?

• What did the alternative sensitivity calculations between approximately 3.0 and 3.7 change, and what were their actual numerators and denominators?

What can a purchaser rely on?

The July 2024 study improves on the earlier analysis by attempting to incorporate the employer's full investment in Lyra. It also demonstrates a substantial observed claims difference between matched users and nonusers.

But the published report does not provide enough information to independently reproduce the reported 3.04 ratio from its disclosed figures, and reconciling the denominator would not by itself establish that the full $4,138 matched claims difference was caused by Lyra.

The 3.04 is therefore useful evidence of how Lyra and the study translate the observed claims difference into an economic return. Before that return becomes an underwriting assumption, a purchaser would need a reproducible calculation and greater confidence that the savings numerator represents costs avoided because of Lyra rather than differences that would have existed without it.