| Study | Aim | Setting and sample period | Sample size and selection | Method | Accuracy measures | Statistical analysis | Results |
| Langley et al,5 2006 | To determine level of coding accuracy for injury PDx and external cause code in hospital discharges | New Zealand public hospitals 1996–1998 | 1670 cases from 52 hospitals; simple random sample of cases with injury as PDx | Independent expert coder recoded medical records mostly blinded to original codes | Accuracy categories:1. Correct2. Correct to 4th digit3. Correct to 3rd digit4. Correct to group level5. Incorrect group level | % correct;Logistic regression:IV, size of hospitalControlled, injury PDxDV, correct/incorrect | 82% external cause coded data overall correct to 4 digit level (18% error)Unintentional % correct:73% Falls75% Other injuries81% MVTCIntentional % correct86% Assault83% SuicideLog reg: No difference in level of coding accuracy by size of hospital |
| LeMier et al,7 2001 | Evaluate accuracy of external cause-of-injury codes in hospital discharges | Washington State (USA) civilian hospitals from 1996 | 1260 cases from 32 hospitals; 32 hospitals selected which accounted for 80% of injury hospitalisations. Simple random sample of cases with injury as PDx | Independent expert coder recoded medical records blinded to original codes | Accuracy categories:1. Correct2. Correct to 4th digit3. Correct to 3rd digit4. Correct to group level5. Incorrect group level | Estimated counts and incidence ratios by mechanism,sensitivity and PPV, inter-rater agreement and kappa.Calculations used sample weight and population corrections where appropriate | Correct ranking of leading causes of injury.Incident rate ratios:20% of other/unspec could have had mechanism coded;48% under-reporting of undetermined intent miscoded as unintentional.Agreement:87% for mechanism95% for intent66% for complete e-code (34% error).Sensitivity and PPV:95% and 93% for falls97% and 88% for MVTC99% and 94% for poisonings92% and 94% for firearms |
| MacIntyre et al,8 1997 | Ascertain the reliability of injury data in hospital discharges | Victoria (Australia) public hospitals 1994–1995 | 546 cases from 4 hospitals; 4 hospitals selected which accounted for 25% of injury hospitalisations. Random sample of cases with injury as PDx and an external cause coded | Medical records reviewed and recoded by doctor with knowledge of ICD-9-CM coding, with input from coder | “Consistency” or “discrepancy” between original codes and medical record content. Categories of “error”:1. Errors of omission2. Superfluous coding3. Discrepancy | Overall discrepancy rates; Diagnosis-specific discrepancy rates;Logistic regression:IV, emergency admission, LOS, No of diagnoses and procedures, type of injury, patient age, hospital, mortality.DV, error in e-code assigned | 84% consistent external cause code assignment (16% error).Types of errors:21% error of omission11% superfluous e-codes68% discrepant e-codes.Log reg: No significant predictor of error in e-codes |
| Langlois et al,6 1995 | Evaluate the quality and availability of cause-of-injury information in medical records and extent to which inadequate documentation contributes to incomplete e-coding | Rhode Island (USA) hospitals 1988–1990 | 1440 cases from unstated number of hospitals. Stratified random sample (based on specificity of e-code) of cases with injury as PDx from all hospitals | Medical records reviewed and recoded by two independent expert coders. Two researchers reviewed codes and assigned final e-code, and cases of disagreement were arbitrated by the first author.Narrative description of cause-of-injury, place of occurrence, and details regarding which health professional recorded the information (eg, doctor, nurse) recorded separately for each form in record (eg, ambulance form, ED form, history) | Level of agreement of original codes and recoded data. Agreement levels:1. Complete agreement to the 4th digit2. Agreement to the 3rd digit3. Agreement to the section4. Disagreement.level of documentation from specific forms assessed by comparing codes assigned from each form to codes assigned for complete record | % agreement | 82% agreement for external cause codes where a specific external cause code assigned originally. Of 18% cases where disagreement present, 64% due to difference in intent.70% of cases originally coded with a vague/unspecified e-code and 66% of cases where no e-code assigned originally were assigned a specific e-code on recode.Medical record documentation was sufficient to assign a specific e-code in 78% of cases and adequate for a vague e-code assignment in an additional 10% of cases.% of cases with specific cause information in different forms:51% History form42% ED record40% Discharge summary |
| Smith et al,10 1990 | Evaluate the usefulness of ICD external cause codes for injury surveillance | Indian Health Service (USA) hospitals from 1985 | 323 cases from 2 hospitals | Medical records reviewed, cause-of-injury information abstracted, and external cause recoded by an independent coder who was blinded to codes originally assigned | Agreement levels:1. Complete agreement to the 3rd and 4th digit2. Agreement to external cause group level3. Agreement by intent | % agreement. | 63% agreement to the 3 digit ICD code level.81% agreement to the external cause code group level.86% agreement a the level of intent. |
DV, dependent variable; ED, emergency department; IV, independent variable; LOS, length of stay; MVTC, motor vehicle traffic crash; PDx, principal diagnosis; PPV, positive predictive value.