15_TCRS_Sec_J_Atch_J-5_DHA_Sampling_Meth.pdf

Attachment J-5 DHA Sampling Methodology

TCRS HT9402-17-R-0002, Section J, Attachment J-5

Department of Defense – Defense Health Agency TRICARE Health Benefits Program

Sampling Methodology

TCRS HT9402-17-R-0002, Section J, Attachment J-5 Page 2 of 6

Department of Defense – Defense Health Agency Military Healthcare Benefits Program

Sampling Methodology

The TRICARE Claims Review Services (TCRS) compliance reviews (i.e. audits) include three sample types: a paid sample (to ensure payment accuracy by identifying underpayments and overpayments), a denied sample (to ensure proper claim denial), and an occurrence sample (to ensure the accurate reporting of healthcare data, regardless of payment accuracy). Paid samples are conducted as a stratified random sample based on paid amounts; denied samples are conducted as a stratified random sample based on billed amounts; and occurrence samples are conducted as a simple random sample with no stratification. The Defense Health Agency (DHA) continually evaluates the accuracy and design of its sampling methodologies for all contracts and implements revisions should they be warranted by audit universe distributions or the outcome of audit results.

Respective to TRICARE purchased-care contract requirements; samples are drawn on either a quarterly or semi-annual basis and include underwritten and non-underwritten healthcare costs.

In addition to quarterly and semi-annual audits, annual health care cost (AHCC) audits are performed on the Managed Care Support (MCS) contracts to determine the total overpayment amount for underwritten healthcare costs that is recovered at the end of each option period;

AHCC audits do not include claims for non-underwritten healthcare costs. Reference TCRS contract Section C, paragraph 1.1.1 for TRICARE purchased-care contract definitions.

• Non-Underwritten healthcare costs are defined as costs that are incurred when the TRICARE purchased-care contractor acts as a Fiscal Intermediary for the Government to distribute, or pass-through, Government funds for certain non-underwritten health care benefits. These are not costs to the TRICARE purchased-care contractor and are not costs that are reimbursed to the TRICARE purchased-care contractor(s) by the Government.

• Underwritten healthcare costs are defined as costs that represent covered care unless specifically excluded by the managed care contracts. The underwritten mechanism consists of an underwriting fee which may be considered to be an underwritten premium associated with the risk assumed by the MCS contractor(s). There is potential for the MCS contractor(s) to earn a negative fee if the actual healthcare costs for a given contract year are significantly higher than a specified target cost for that year.

Universe and Sample Payment Data

DHA payments subject to sampling are recorded as TRICARE Data Encounter (TED) records.

Each TED record serves as an invoice to DHA for care paid by TRICARE purchased-care contractors. DHA stores all TED data in a central repository, the TED Operational Data Store

(TEDODS).

DHA relies on the TED Audit (TA) application throughout the sampling process. The number of strata, lower and upper boundary of each stratum, as well as stratum sample sizes are stored in

TCRS HT9402-17-R-0002, Section J, Attachment J-5 Page 3 of 6

TA for each audit. TA pulls TED records from the TEDODS to build the audit universe and randomly selects the appropriate number of claims to generate each sample. Before a random sample is drawn, the audit universe constructed by TA is validated against data queried independently from the TEDODS. The number of records as well as the total paid and billed amounts are verified to ensure that all payments subject to sampling are available in the universe from which claims are randomly selected to be audited.

One exception to this process is the Active Duty Dental Program (ADDP) dental data, which does not produce TED records. Alternatively, TA obtains ADDP transactions from DHA’s Military Health System (MHS) Data Repository (MDR). ADDP universes are validated by comparing TA’s data with raw data extracts from the MDR before samples are drawn.

Non-Denied Payment Samples

Overview Non-Denied payment samples are conducted to identify improper payments and measure payment accuracy. DHA has chosen to use stratified random sampling to identify the most improper payments at the lowest cost.

Depending on the TRICARE contract type, the universe for a paid sample may contain between several hundred thousand to 30 million claims. All claims with a government payment amount above a high-dollar threshold (i.e., $200,000) are reviewed, and claims below this threshold are randomly sampled based on stratification of the government payment amount. The high-dollar threshold may fluctuate from one audit to the next, depending on the composition of the claims payment universe. Low-dollar thresholds (i.e. claims with government payment amounts < $100.00) have been established in some contracts for sample exclusion due to historically low error rates for these claims and to ensure efficient utilization of audit resources.

Stratification Samples for paid claims include between four and 12 strata, depending on the composition of the claims in the universe. The distribution of claims payments is heavily positively skewed, with the majority of claims payments falling below $1,000 and others being as high as millions of dollars. Visual analysis of the distribution of payments shows no natural strata separation (see Figure 1 below for an illustrative example).

Figure 1 – Typical Distribution of Paid Amounts

TCRS HT9402-17-R-0002, Section J, Attachment J-5 Page 4 of 6

Since strata cannot be easily identified via visual analysis of payment distributions, mathematical methods assist in the identification of optimal strata boundaries. The Cumulative Square Root Frequency (CSRF) method is effective for identifying strata boundaries in heavily skewed distributions and is often used in healthcare claims audits1. The CSRF method can optimally stratify the universe of records (by payment amount) to reduce variability within each strata and achieve the most efficient sample design. DHA has simulated thousands of samples and determined that the CSRF method is the most effective in delineating strata for our sampling needs.

Strata boundaries identified by the CSRF method are rounded slightly to allow generalization and avoid over-fitting to one sample universe. For example, a boundary identified to be $12,519.76 may be rounded to $12,520.00; or a boundary calculated at $237,904.12 may be rounded to $237,905.00. This promotes a cleaner sample design.

Sample Size Calculation Once the number of strata is chosen and strata boundaries are identified, the sample size for each individual stratum is calculated via Formula 1 (below) to yield an estimate with a minimum of 90% confidence plus or minus 2.5 percentage points. If additional resources are available, these parameters may be made stricter (up to 99% confidence and/or plus or minus as little as 0.5 percentage point) to be conservative and result in a larger sample size, which would in turn increase the likelihood that the sample meets precision targets.

Formula 12: 𝒏 = 𝑧2𝑁𝑉𝑥2 𝑧2𝑉𝑥2+(𝑁−1)𝜀2

Where: 𝑛 = sample size 𝑧 = reliability coefficient (1.96 for a 95% confidence level) 𝑁 = population size

𝑉𝑥2 = relative variance for the variable 𝑥, calculated as 𝜎𝑥2

�̅�2 𝜀 = acceptable error amount expressed as a percentage (i.e., 5%)

The MCS AHCC audits include up to 10,000 underwritten healthcare claims. This sample size is larger than the result of Formula 1 (above) to be conservative and provide additional assurance that precision targets are met, which is important since results from AHCC audits are extrapolated to determine the total overpayment amount of the claims universe that is recovered from the contractor (discussed in further detail below).

Case Selection Probability

1 Buddhakulsomsiri, J., Parthanadee, P., Kachhal, S. “A Stratified Sampling Plan for Billing Accuracy in Healthcare Systems.” Department of Industrial and Manufacturing Systems Engineering, University of Michigan-Dearborn;

Department of Agro-Industry Technology Management, Kasetsart University, Thailand.

http://www.iienet2.org/uploadedfiles/SHSNew/Tools_and_Resources/Sampling%20Plan%20for%20Billing%20Acc uracy%20Paper.pdf 2 Formula 3.14, page 62; Sampling of Populations: Methods and Applications; Levy, P. S. & Lemeshow, S. (1991);

New York: John Wiley & Sons.

TCRS HT9402-17-R-0002, Section J, Attachment J-5 Page 5 of 6

Once claims are stratified and sample sizes are calculated, claims for each stratum are selected via simple random sample. Claims are selected with equal probability. Again, 100% of claims are selected in the stratum covering high-dollar claims.

Denied Payment Samples

Overview The primary purpose of the denied payment sample is to ensure that health care/supplies are not being denied inappropriately by contractors.

The TED records that encompass the denied payment sample universe include records with government payment amounts equal to zero as a result of claim cancellation, denial of healthcare services, or claims with zero payment due to 100% payment by Other Health Insurance (OHI).

DHA has chosen to use stratified random sampling to identify the most improper payments at the lowest cost.

Stratification The denied payment sample is similar in design to the payment sample; the primary difference is that the denied sample is stratified based on billed amount since the paid amount for a denied claim will be $0. Depending on the contract type, a denied audit universe may contain between several thousand to more than 1 million claims. All claims with billed amounts above a high-dollar threshold (i.e., billed amounts > = $200,000) are audited, and similar to the design of the payment sample, a low-dollar threshold may also be implemented. These thresholds may fluctuate across audit cycles and are dependent on the composition of claims in the audit universe. Claims between the thresholds are stratified on the billed amount, and samples include between four and 12 strata. Similar to paid amounts in the payment sample, the distribution of billed amounts is always heavily positively skewed, and the CSRF method is used to identify strata boundaries.

Sample Size Calculation Once the number of strata is chosen and strata boundaries are identified, the sample size for each individual stratum is calculated to yield an estimate with a minimum of 90% confidence plus or minus up to 7.5 percentage points (via Formula 1). If additional resources are available, these parameters may be made stricter to follow a conservative approach that results in a larger sample size, which would provide additional assurance that precision targets are met. DHA understands that 90% confidence and 15% precision do not conform to IPIA guidance, but the denied claims have zero government liability, so the sampling parameters required to measure payment errors are unjustified. Again, the purpose of the denied sample is to ensure that claims are being denied appropriately, not to measure the accuracy of paid amounts.

Case Selection Probability Similar to the design of the payment sample, claims for each stratum are selected via simple random sample, with equal probability. Again, 100% of claims are selected in the stratum covering high-dollar claims.

Occurrence Samples

TCRS HT9402-17-R-0002, Section J, Attachment J-5 Page 6 of 6

Overview The occurrence sample is intended to monitor and evaluate the accuracy of TED record coding by the TRICARE purchased-care contractors, as opposed to identifying payment errors.

Accordingly, these records are selected via a simple random sample rather than a stratified random sample. A flat sample size (of up to 350 records) is selected for each occurrence sample, and each record in the sample contains approximately 90+ data fields that are reviewed for accuracy. Records for the occurrence sample are selected with equal probability.