Colorado claimed $8M in Medicaid payments it can't verify

Every state must electronically verify that a Medicaid home-care attendant actually showed up -- an Electronic Visit Verification (EVV) check-in by phone or app, timestamped and geolocated.▸ A July 2026 HHS Office of Inspector General audit⧉ tested whether Colorado's system was working. It sampled 160 personal-care claim lines from a pool of 1,301,307 claim lines worth $299.2 million ($152.9 million federal share) paid during the year running July 2023 through June 2024. Ninety-one of the 160 -- 57% -- had at least one visit that couldn't be electronically verified. Projected across the full year, OIG estimates 52.5% of Colorado's personal-care claim lines failed verification, and recommends the state refund $8.07 million in federal reimbursement for services that didn't meet billing rules, with a further $45.7 million now sitting in front of CMS pending a decision.

Colorado's EVV system lets an attendant check in four ways: a mobile app, a phone call, a manual entry by the provider, or another approved method.▸ Of the 91 sampled claim lines with a compliance problem, 62 involved at least one visit entered manually rather than captured in real time -- for one enrollee, every single EVV record for a full year, 185 of them, was a manual entry. Another 17 claim lines had visits never entered into the system at all: one covered 29 dates of service with EVV records for only 6. Sixteen had no documented service location, and 16 more had GPS data placing the attendant away from the enrollee's home at check-in or check-out that nobody ever reviewed or corrected -- in one case, all 8 recorded visits for a claim line lacked check-in GPS data entirely, with 6 of 8 check-outs flagged as off-site and no alternate location ever logged.

A separate question -- whether the billing itself was proper -- turned up 39 sampled claim lines that were at least partially unallowable.▸ The largest cause, in 23 of the 39, was a missing or incomplete background-screening record for the attendant. Fourteen involved units of service that exceeded what the attendant's own timesheet documented -- most strikingly one claim line where a provider billed 3,876 units (about 121 hours a day) over 8 dates, while the timesheets for those same dates supported only 161 units, about 5 hours a day; the provider blamed a backlog from delayed eligibility renewals. Eight claims billed Colorado's enhanced Denver County rate for services delivered -- and geolocated -- outside Denver County, including one enrollee roughly 50 miles away in Weld County. Four lacked an approved care plan altogether.

OIG's $8.07 million refund recommendation is deliberately conservative -- the low end of a 90% confidence interval designed to understate the true total at least 95% of the time; the point estimate is $16.2 million.▸ But a second, larger problem surfaced in the 50 sampled claim lines billed as consumer-directed care, where enrollees hire and manage their own attendants: every single one of those timesheets recorded only a date, a time in and out, and the label "CDASS," with no detail on which specific tasks -- bathing, dressing, meal prep -- were actually performed. Because none of the 50 could be checked against the enrollee's approved care plan, OIG set aside an estimated $45.7 million in federal reimbursement (point estimate) for CMS to resolve. Combined, the $8.07 million recommended refund and the $45.7 million set-aside equal $53.8 million -- 35% of the entire federal share Colorado billed for personal care that year.

Colorado's Department of Health Care Policy and Financing disagreed with most of OIG's findings, and its central objection was methodological: it argued the sample wasn't statistically representative, pointing out that 6 of the partially unallowable claim lines traced to a single enrollee and that 13 of 14 excessive-unit findings traced to just 3 provider agencies -- all concentrated in one of the audit's four sampling strata.▸ The state also cited its own statewide figures showing 19-25% of visits are manually entered, versus the 38% manual-entry rate in OIG's sample, arguing the sample doesn't reflect typical statewide practice. OIG rejected the representativeness challenge. It ran a properly executed stratified random sample using RAT-STATS statistical software, the kind federal courts have repeatedly upheld for Medicare and Medicaid overpayment extrapolation -- including samples smaller than the 160 items used here. Design-based sampling, OIG noted, doesn't require the sample's items to be independent of each other; clustering inside one stratum doesn't contaminate the estimate for the rest.

Colorado separately argued that gaps in decade-old background-screening records reflect retention policy, not proof that any attendant was actually ineligible or that care wasn't delivered.▸ OIG's answer: without the documentation, the state cannot show the payments complied with the rules it agreed to follow, whatever the underlying reason for the gap.