6 Claims Lag and Runout
A sudden drop in recent medical spending is often the first exciting result in a new claims dataset. It is also one of the first results to distrust. The care may already have happened while the claims are still being submitted, processed, paid, or delivered to your warehouse.
This delay is claims lag. It affects cost, utilization, diagnoses, and any other analysis that depends on claims becoming visible. In this chapter, we’ll distinguish the dates involved, learn to compare periods fairly, and work through a simple completion estimate. The numbers are illustrative; they are not benchmarks for how quickly any particular payer’s claims develop.
6.1 Three clocks to keep separate
Claims analysis usually involves three different timelines.
The incurred period assigns the service or financial obligation to a period of analysis. For a medical utilization report, this is commonly based on service dates. A hospital stay crossing month-end requires a convention: admission month, discharge month, service-line dates, or an allocation across days. The choice should follow the measure. There is no universal incurred_date field that removes this decision.
The paid period records when the payer paid the claim or posted the relevant payment transaction. An adjudication date records when the claim was processed. These can differ. If a feed supplies a processed date as a proxy for paid date, document that substitution and use it consistently. A processed-date analysis cannot automatically be interpreted as a cash-payment analysis.
The data as-of date describes the information available for the analysis. A file received in July might contain payments only through June. Another file received on the same day might be missing a week of June activity. Record both the source cutoff and the delivery used; the warehouse load date alone does not establish completeness.
For example, “January–June services, paid through June 30, using the July 10 extract” is much more informative than “first-half claims.” See Key Data Elements for the underlying dates.
For an incurred cohort, an incurred claims estimate includes payments to date plus estimated unpaid obligations. This differs from incurred expense for an accounting period, which includes payments during that period plus the change in unpaid claims liabilities from its beginning to its end. The Actuarial Standards Board’s ASOP No. 5 distinguishes these definitions and defines the valuation date. Be clear about whether a report shows observed spending, estimated ultimate cost for a cohort, or accounting-period expense.
6.2 Why recent periods are incomplete
A service can pass through several organizations before it appears in an analytical extract. The provider must finish the documentation and billing. A clearinghouse may return an error. The payer may need additional information. A payment can then wait for the next extract and delivery cycle.
This means a complete file of payments made in June is not a complete record of services performed in June. Some June services will appear in July or later. June’s file may also contain payments for services from much earlier months.
Different services and sources can have different development patterns. A quick pharmacy transaction and a disputed hospital bill should not be assumed to arrive at the same speed. Nor should an analyst receiving a delayed monthly extract assume the same visibility as the payer’s internal team.
Lag affects denominators and numerators differently. Enrollment may arrive before the corresponding claims, making recent PMPM or utilization rates appear low. Coverage can also change retroactively. Review both sides of a rate before attributing a change to clinical performance.
6.3 Runout: allowing time for claims to arrive
Runout is the additional period during which claims for an earlier incurred period continue to be received or paid. Under a monthly convention, January claims observed through the end of January have age 0; observed through the end of February, they have one month of runout, or age 1.
There is no fixed amount of runout that guarantees completeness. Select a cutoff using the source’s observed development, the service category, the size of the remaining uncertainty, and any applicable contractual specification. Three or six months may be a reporting convention, but neither is evidence by itself that every claim has arrived.
One practical approach is to compare periods at equal maturity. Compare January through March this year with January through March last year, giving both periods the same number of subsequent paid months. Otherwise, a mature prior year compared with an immature current year can create apparent savings that disappear as claims arrive.
Waiting improves completeness at the cost of timeliness. When that tradeoff does not fit the question, we can estimate the missing amount. The estimate needs to remain distinguishable from the observed data.
6.4 Reading a development triangle
A development triangle, also called a lag triangle, organizes claims by incurred period and development age. The following table shows cumulative net paid amounts, in thousands of dollars, as of June 30 of an illustrative year.
| Incurred month | Age 0 | Age 1 | Age 2 | Age 3 | Age 4 | Age 5 |
|---|---|---|---|---|---|---|
| January | 60.0 | 85.0 | 95.0 | 98.0 | 99.0 | 100.0 |
| February | 72.0 | 102.0 | 114.0 | 117.6 | 118.8 | — |
| March | 54.0 | 76.5 | 85.5 | 88.2 | — | — |
| April | 66.0 | 93.5 | 104.5 | — | — | — |
| May | 60.0 | 85.0 | — | — | — | — |
| June | 78.0 | — | — | — | — | — |
Read across the January row: $60,000 was paid during January for January services, and $85,000 had been paid by the end of February. The additional payment during February was therefore $25,000. The table is cumulative, so adding cells across the row would double-count payments.
Read down the age 1 column to compare different incurred months at the same stage of development. Read the last observed cell in each row to see what is available at the common June 30 cutoff. The blank cells are future observations, not zero payments. Their shape gives the triangle its name.
Build the triangle from a consistent financial basis and account for adjustments and reversals. With recoveries or reversals, cumulative net paid amounts can decrease. Do not force the data to increase merely because a simple teaching example does.
A current final-state claims table may not retain what was known at earlier cutoffs. Reconstructing that history requires suitable transaction dates and amounts, historical snapshots, or another source that preserves development. Moving every adjustment back to an original payment date can make a historical triangle look more complete than it actually was.
6.5 From development to completion
To estimate ultimate cost, we need an assumption about what share has been observed at each age. Call this the completion percentage, \(c_a\), for age \(a\).
One simple way to estimate it is to use older, sufficiently mature cohorts and divide their cumulative paid amount at a selected age by their selected ultimate amount. For example, suppose three older cohorts had paid $850,000, $1,020,000, and $680,000 by age 1, with selected ultimate amounts of $1 million, $1.2 million, and $800,000. Their combined age 1 completion percentage is:
\[ c_1 = \frac{850{,}000 + 1{,}020{,}000 + 680{,}000} {1{,}000{,}000 + 1{,}200{,}000 + 800{,}000} = 0.85. \]
This ratio weights cohorts by ultimate dollars. It is not generally the same as a simple average of monthly percentages. The ultimate amounts are themselves selected estimates unless the cohorts are considered fully settled; historical data does not make the ultimate automatically known.
For observed paid amount \(P_a\) and a positive selected completion percentage \(c_a\):
\[ \widehat{U} = \frac{P_a}{c_a}, \qquad \widehat{\text{unpaid}} = \widehat{U} - P_a. \]
Some teams call \(c_a\) a completion factor. Others use “factor” to mean the reciprocal development multiplier, \(1/c_a\). Write down the formula to remove the ambiguity: 85% completion means dividing by 0.85, or multiplying by approximately 1.1765. Multiplying by 0.85 would reduce an already incomplete amount.
Assume a separate analysis of mature history led us to select completion percentages of 60%, 85%, 95%, 98%, 99%, and 100% for ages 0 through 5. Applying these illustrative selections to the triangle gives:
| Incurred month | Age at June 30 | Observed paid | Selected completion | Estimated ultimate |
|---|---|---|---|---|
| January | 5 | $100,000 | 100% | $100,000 |
| February | 4 | $118,800 | 99% | $120,000 |
| March | 3 | $88,200 | 98% | $90,000 |
| April | 2 | $104,500 | 95% | $110,000 |
| May | 1 | $85,000 | 85% | $100,000 |
| June | 0 | $78,000 | 60% | $130,000 |
| Total | $574,500 | $650,000 |
The estimated unpaid amount is $75,500. The assumption of 100% at age 5 is only for this example; real selections may allow development well beyond that point. These calculations illustrate the mechanics of a completion estimate, not how to choose a reserving method for a particular population.
6.6 What the estimate can and cannot tell us
The biggest uncertainty is usually in the newest months. With $78,000 observed for June, selecting 50% completion yields $156,000; selecting 70% yields about $111,429. A plausible-looking factor can materially change the result.
Payment patterns can also change. A new claims system, missing delivery, processing backlog, or change in the mix of services can make older experience a poor guide. Investigate these events before applying historical factors. Segmenting by service or payer can help, but very small groups may not provide enough stable experience for separate estimates.
Dollar completion is not necessarily encounter completion. A missing high-cost admission can affect spending much more than visit counts. Use a development approach appropriate to the measure instead of applying a cost factor to every claims-derived metric.
Keep observed and estimated amounts side by side, state the cutoff and selected assumptions, and compare prior estimates with later results. Building the warehouse inputs and applying supplied factors are data transformations. Selecting and evaluating estimates of unpaid liabilities is actuarial work; ASOP No. 5 addresses that broader judgment and follow-up. A completed claims chart should help readers see uncertainty, not make it disappear.