APPENDIX: Detailed Methods

November/December 2012 Vol 5, No 7

National Burden of Preventable Adverse Drug Events Associated with Inpatient Injectable Medications


Data Sources

Five databases were used to estimate the healthcare costs associated with inpatient medication errors, and 3 sources were used for medical professional liability(MPL) costs, as described in the Table. A study by the Office of Inspector General of the US Department of Health and Human Services (HHS) was the source for the per-admission frequency of avoidable inpatient medication errors and incremental inpatient cost.1 A medical liability industry rate survey2 was the source for relative MPL costs by state.

Table Databases Used to Estimate Costs Associated with Medication Errors

Database (owner)


Sample size (yrs)

Use in study

Databases used for healthcare costs

Quantros MedMarx (Quantros, Inc, Marlborough, MA)

Self-reported medication errors by medication; participating hospitals

166,498 events (2009-2011)

Identification of particular medications associated with medication errors

The Premier Database-Premier Research Services (Premier, Inc, Charlotte, NC)

Patient-level detail of procedures, diagnosis, DRG, and medications in the inpatient setting; participating hospitals

5.6 million discharges (2010 and 2011)

Per-patient use of particular medications by patient DRG

Thomson Reuters MarketScan Research Databases (Truven Health Analytics, Ann Arbor, MI)

Commercial insurer administrative data (claims and exposure); contributing payers

40.0 million members in 2009 and 45.2 million members in 2010 (2009 and 2010)

Incremental cost associated with reported ICD-9s indicating inpatient medication errors in commercial population

Medicare 5% Analytic Sample (Centers for Medicare & Medicaid Services)

Medicare fee-for-service insurer administrative data (claims and exposure)

2.4 million beneficiaries in 2009 and 2.5 million beneficiaries in 2010 (2009 and 2010)

Incremental cost associated with reported ICD-9s indicating inpatient medication errors in Medicare population

State Discharge Database (Milliman assembled)

All payer databases of all discharges by DRG in 17 states (AZ, CA, FL, IA, IL, MA, MD, NJ, NY, OK, RI, TX, UT, VA, VT, WA, WI)

20.7 million discharges (2010)

Standard per-capita distribution of DRGs

Databases used for MPL costs

National Practitioner Data Bank (US Department of Health and Human Services)

MPL claims reported by state licensing authorities

864,702 claims (1990-2011)

Size and frequency of claims for inpatient medication errors relative to all MPL claims

MPL insurance filings for states (various MPL insurance companies)

MPL premium rate development filed with state insurance regulators, publicly available

8 filings: CA, FL, LA, MA, NC, OH, PA, VT (2007-2011)

Premium rates for facility portion of MPL

American Hospital Association Annual Survey (American Hospital Association, Chicago, IL)

Operational and financial statistics on individual hospitals

6334 hospitals (2010)

Number of beds by region and bed type, occupancy rates, births, and inpatient procedures

Closed Claim Database (Florida Department of Insurance)

Chapter 627.912, F.S. of Florida Insurance law requires insurance companies, self-insurance funds, and joint underwriting associations to file claim reports for insured entities and individuals

64,469 defendants (1994-2009)

Inpatient professional MPL claim cost for inpatient cases relative to hospital MPL cost for inpatient cases

DRG indicates diagnosis-related group; ICD-9, International Classification of Diseases, Ninth Revision; MPL, medical professional liability.



For medical costs, the conditional probability of an inpatient medication error for each type of injectable medication in the MedMarx database was developed. The inpatient use rate of each medication type by diagnosis-related group (DRG) was then applied to the conditional probability to develop the probability of an inpatient medication error by DRG, and then the number of errors per inpatient discharge was expanded by applying the annual number of US inpatient discharges. Costs per inpatient medication error were estimated and then applied to the number of inpatient errors to develop an annual national cost estimate.

For MPL costs, the portion of inpatient MPL associated with medication errors was estimated, and that ratio was applied to the estimated per-bed MPL costs for inpatient hospital exposure. By applying the proportional contribution of professional liability for inpatient care to facility liability, the MPL estimate includes both the facility and professional liabilities for all nongovernment hospital beds in the United States.

Overview of Methodology for Medical Costs

Using numbers from the HHS study, the probability that an inpatient stay will experience a medication error is the probability of an adverse event (26.9%) multiplied by the proportion of adverse events that were medication related (37%), which produces a 10.07% rate of adverse drug events (ADEs). The proportion of adverse events that were medication related includes both “temporary harm” and “permanent harm” events, as defined in the HHS study, and the cost estimate includes both of these categories, as described below. Of the total ADEs, 50% are estimated to be avoidable. We spread the resulting avoidable ADEs to particular injectable medications and particular admission types by DRG.

The distribution of injectable medications, given that an inpatient medication error has occurred, was estimated from MedMarx, as was the portion of inpatient medication errors resulting from injectable medication administrations (87%). Using Premier, the distribution of injectable medication administrations by type of medication for each DRG was developed.

The probability of having an avoidable ADE given an administration of an injectable medication was calculated using Bayes’ theorem:

  • P(InjRx|ADE): Portion of the inpatient admissions related to each injectable medication among avoidable ADE inpatient admissions from MedMarx
  • F(InjRx): Frequency of administration of each injectable medication per inpatient admission from Premier
  • P(ADE): Probability of avoidable ADE resulting from an injectable medication occurring per inpatient admission, from the HHS study (10.07% × 50% × 87%).

Costs per adverse event were estimated from MarketScan and the Medicare 5% Sample as the incremental cost of patients with diagnosis codes indicating inpatient medication errors relative to matched patients without such codes.

Calculations Using Premier Data

The number of administrations of injectable medications per inpatient admission in the Premier data was summarized by injectable medication name according to DRGs from Premier. The frequency was calculated as the number of injections per inpatient admission, which was, on average, 15.39 across all inpatients. Any medication with fewer than 0.01 administrations per admission (across all admissions) was excluded from consideration.

Calculations Using MedMarx Data

The distribution of inpatient injectable medication errors by medication was developed from MedMarx data. Injectable inpatient ADE cases in MedMarx were those with error category E (ie, error, harm) or worse. The total number of injectable ADE cases in MedMarx was 3693. The proportion of cases related to each injectable medication was calculated as the number of ADE cases related to the injectable medication divided by the total number of ADE cases in MedMarx. Based on the HHS study, the proportion of avoidable and unavoidable cases related to each injectable medication was assumed to be 50%.

Probability of Avoidable ADE per Administration of Injectable Medication

Using Bayes’ theorem, the probability of avoidable ADE per administration of injectable medication by DRG was calculated as the probability of an injectable medication given an avoidable ADE divided by the administration frequency of the injectable medication per DRG. The aggregate probability of avoidable ADEs was calibrated to match the estimate from the HHS study (3.8%), adjusted by the 87% of inpatient medication errors attributed to injectable medications derived from MedMarx.

The probability of avoidable ADE per administration of a given injectable medication was assumed to be the same for each DRG. The incidence of avoidable ADEs was assumed to follow a binomial distribution with respect to the number of administrations—that the probability of an error for each administration is independent of whether there were any previous errors.

Incremental Cost Calculation Using MarketScan and Medicare 5% Sample

To estimate the incremental costs of ADE cases, we selected International Classification of Diseases, Ninth Revision(ICD-9) diagnosis codes that indicated medication errors among discharges that were assigned surgical DRGs. Surgical DRGs were chosen because clinical review suggested that, with surgical cases, the selected ICD-9 diagnosis codes would identify errors occurring during the inpatient stay rather than before the inpatient stay.

The HHS study had identified incremental costs as the impact on DRG coding of removing procedure and diagnosis codes associated with the error; however, our case identification methodology based on surgical DRGs did not produce any impact on DRG coding when the diagnosis codes identifying errors were removed. Therefore, we calculated that the incremental cost of avoidable ADEs was calculated by payer as the sum of 2 pieces: the incremental cost of changes to the DRG based on the HHS study and the non-DRG incremental cost (including incremental costs after the ADE admission). The second piece, the non-DRG incremental cost, was tabulated as the difference between the costs of reported ADE claims and of matched, reported non-ADE claims, both within 4 months of the non-ADE admission date.

To find matched cases, we chose non-ADE claims whose costs during the 3 months before the discharge were within 3% of costs of ADE claims. This was performed separately for MarketScan and for the Medicare 5% Sample. Differences were tabulated for the 4 months after admission.

Total Incremental Cost of Avoidable ADEs

The distribution of inpatient admissions by payer and by DRG was derived from 2010 data from the state discharge data of 17 states. We extrapolated the total discharges from these 17 states to match the US total admissions of approximately 37 million.3

The total incremental cost of avoidable ADEs by payer by DRG was calculated as the number of inpatient admissions by DRG multiplied by the probability of avoidable ADE given a DRG multiplied by the incremental cost per avoidable ADE. The probability of avoidable ADEs for a given DRG was assumed not to vary by payer and the incremental cost of an avoidable ADE by payer was assumed not to vary by DRG.

Overview of Methodology for Medical Professional Liability Costs

Nationwide hospital MPL costs were first estimated by region as the product of the facility MPL rate per hospital bed and the number of beds in the region. We calculated a national estimate based on the sum of regional estimates. We multiplied this sum by the portion of the MPL premium rate attributed to inpatient medication errors, a multiplier to account for professional MPL costs, and a multiplier to account for the portion resulting from injectable medications.

Calculations Using MPL Premium Rates

The annual premium rate for hospital liability costs was estimated by analyzing publicly available rate filings that were submitted to insurance regulators by MPL insurance market leaders in several states. Premium rates charged by MPL insurers include MPL settlement costs, legal expenses, administrative expenses, and insurer profit. Although the details vary by company and regulatory requirements, premium rate filings typically identify the annual MPL premium charged to a hospital by bed. Rates typically vary for a hospital by type of bed (eg, maternity, surgery), type of procedure, and other factors. Using the information in rate filings, premium rates were calculated that corresponded only to claims related to inpatient care, such as inpatient beds and certain inpatient procedures (eg, births), excluding any portion of premium attributed to outpatient services, hospital-employed physicians, or nursing facilities. This yielded the annual MPL premium rate for hospital coverage per bed (or other category, such as births and inpatient procedures) for a regionBed counts (adjusted for occupancy), births, and inpatient surgical procedures from the 2010 American Hospital Association Annual Survey were used to estimate inpatient hospital MPL costs for 8 states, which were chosen because complete rate filings were available in these states and because the states had large populations.

Because recent rate filings were not available in all states, published state MPL cost relativities were applied to the calculated MPL rates from these 8 states. The published relativities were for physician professional liability (PPL) premium rates, because PPL rates are more readily available than hospital liability (HL) rates. This use of PPL relativities assumes PPL and HL costs follow similar patterns among states, which is consistent with assuming a state’s tort environment, court system, and cost level would impact PPL and HL costs similarly.

To estimate the number of HL claims by state related to inpatient medication errors, the average total HL cost per claim, including indemnity, loss adjustment expense, and operating costs, was estimated. The national estimated annual HL cost associated with medication errors was divided by the cost per claim (derived from the National Practitioner Data Bank [NPDB] data, as discussed below) to derive the estimated number of inpatient MPL claims associated with medication errors.

MPL cost estimates are subject to statistical variation. The standard deviation in annual aggregate NPDB medication error claim costs was calculated from claims that were closed over the period of 2004 to 2011. This standard deviation, multiplied by the ratio of the cost estimate for inpatient HL medication errors relative to the NPDB medication error claim costs, was used to estimate the potential variation in aggregate inpatient HL cost resulting from medication errors.

Calculations Using the National Practitioner Data Bank

NPDB data were used to estimate the portion of hospital facility MPL that was associated with inpatient medications. The portion of MPL claim costs associated with inpatient medication errors was calculated as the product of the proportion of claims associated with medication errors and the relative size of medication error claims compared with all MPL claims.

Because the site of medication administration (eg, hospital or community) is unavailable in the NPDB, medication-associated claims associated with nurse activities were used as an approximate way to identify inpatient-associated claims, because nurses are more heavily involved in inpatient medication administration than physicians.

Inpatient Professional Analysis

The MPL cost estimates based on premium rate filings cover claims against hospitals, but they do not cover claims against physicians. To estimate the cost of claims against physicians, a ratio of physician MPL cost to hospital MPL cost was developed using data from the Florida Department of Insurance Closed Claim Database. Although these data are for Florida only, they appear to be the only publicly available sources of MPL cost that allow the separation of costs for inpatient cases. The cost for professionals as a result of inpatient cases is approximately equal to the cost for the hospital. Therefore, the ratio of physician plus hospital MPL to hospital MPL for inpatient medication errors is equal to 2.

Portion of Medication Errors Resulting from Injectable Medications

We used the 87% figure from our MedMarx analysis as the portion of inpatient medication errors that are associated with injectable medications.


  1. Levinson DR. Adverse events in hospitals: national incidence among Medicare beneficiaries. US Dept of Health and Human Services, Office of Inspector General. November 2010. OEI-06-09-00090. Accessed November 15, 2012.
  2. Medical Liability Monitor annual rate survey. Medical Liability Monitor. 2011;36:7-43.
  3. American Hospital Association. Fast facts on US hospitals. Accessed November 19, 2012.
Last modified: July 28, 2015
Copyright © Engage Healthcare Communications, LLC. All rights reserved.