OUP user menu

Identifying risk factors for medical injury

Clare E. Guse, Hongyan Yang, Peter M. Layde
DOI: http://dx.doi.org/10.1093/intqhc/mzi106 203-210 First published online: 9 June 2006

Abstract

Objective. To examine multiple risk factors for medical injury using administrative data.

Design. This cross-sectional study used logistic regression models to examine associations among patient characteristics such as age, sex, and insurance payer status and hospital characteristics such as ownership, teaching status and trauma level, and comorbidities and presence of a medical injury diagnosis. Data were from the Bureau of Health Information: Wisconsin hospital in-patient discharge records and hospital characteristics for the year 2001.

Setting. All Wisconsin non-federal, acute-care hospitals.

Participants. A total of 556 899 patients discharged from 132 Wisconsin hospitals, excluding newborns, participated.

Intervention. None.

Main outcome measure. Medical injury, defined as untoward harm to a patient as a result of a medical intervention, was determined using discharge diagnosis criteria.

Results. Medical injuries were found in 13.3% of 556 899 hospital discharge records. Covariates associated with increased risk of a recorded medical injury code included age 45–84 years, female sex, comorbidities, non-profit religious order ownership, proportion of cardiac intensive care in-patient days to total in-patient days, percent of board-certified medical staff, and community trauma center or regional trauma resource.

Conclusion. This article describes an innovative analysis of risk factors for medical injury that controlled for numerous potential confounding factors, including hospital coding characteristics. The associations we found, such as increased risk of medical injury in women, can be used to generate hypotheses for further testing through other methods and suggest intervention points for patient safety efforts.

  • administrative data
  • medical injury
  • patient safety
  • risk factors

From 3 to 18% of patients discharged from hospitals have experienced medical injuries [1–5]. A few studies have used multivariate regression techniques to examine risk factors for medical injury or error [6–9], but none of these were comprehensive in population studied and risk factors examined. Furthermore, the disparate existing information about risk factors sometimes pertains only to special populations [7–9]. Brennan and colleagues [1] found that adverse event rates increased significantly with age and differed among medical specialties. Burstin and colleagues [6] found that compared with privately insured patients, the uninsured were at increased risk of adverse events because of substandard care but were not at increased risk of adverse events in general. In a comparison of the quality in Australian Health Care Study and Utah/Colorado Study, Thomas and colleagues [10] cite as prior bivariate predictors of adverse events age, payer status, race, hospital teaching status, hospital urban/rural location, and hospital size. In surgical patients, Gawande and colleagues [7] found that emergency surgery, unexpected changes during planned surgery, and patient obesity all contributed to an increased risk of leaving a surgical object in a patient. In a study of children, Slonim and colleagues [8] found associations between medical errors and sex, median household income, special needs or technology dependence, and hospital ownership. Kanter and colleagues [9] found male gender, Medicare/Medicaid and commercial insurance, urban hospitals, and very low or very high birth weights to increase the odds of a medical error in premature neonates.

Most of the studies reviewed above were not undertaken to examine risk factors for medical injury. A better understanding of the inter-relationships of risk factors is needed to facilitate intervention to prevent medical injuries. Our study used multivariate techniques to shed light on the inter-relationships among patient and hospital characteristics and medical injuries that were diagnosed in patients discharged from Wisconsin hospitals, using an administrative data set of hospital discharge records for 2001.

Methods

Data sources

Data for in-patient visits to all Wisconsin non-federal, general, acute-care hospitals from the year 2001 were obtained from the Wisconsin Bureau of Health Information (BHI). The three federal Veterans Affairs hospitals in Wisconsin do not report discharge data to the BHI and were excluded. This data set included one record for each discharge and identified age, sex, primary and secondary insurance payer status, up to nine International Classification of Diseases—9th Revision—Clinical Modification (ICD-9-CM) [11] diagnostic codes and one specific field for an external cause code (E code) for injuries, up to six ICD-9-CM procedure codes, and hospital identification number. Delivery discharges for newborns were excluded, as this group was not included in validation of the screening criteria [12]: their mothers’ data were included.

Data on hospital characteristics obtained from an annual survey of hospitals conducted by the Wisconsin BHI included hospital ownership; number of admissions; discharges and deaths; number of beds and in-patients for selected in-patient units; specialty services such as cardiac intensive care, oncology services, and transplant services; trauma level as designated by the hospital; size of nursing and medical records staff; number of board-certified staff; and number of surgical in-patients and outpatients.

Medical injury screening criteria

The Wisconsin Medical Injury Prevention Program (WMIPP) screening criteria identify medical injuries in routinely collected hospital discharge data based on specific ICD-9-CM discharge diagnoses and E codes, including injuries occurring outside the hospital setting (available at http://www.mcw.edu/medicalinjury [13]). The criteria have been validated and found to be accurate in identifying medical injuries, defined as ‘any untoward harm associated with a therapeutic or diagnostic health care intervention’ [12].

Risk factor definitions

Patient characteristics. The medical injury rate by All Patients Refined—Diagnosis Related Group (APR-DRG) [14] category was calculated using data from year 2000, after removing any injury diagnosis codes. The corresponding rate was then assigned to each patient in the 2001 data according to their APR-DRG. This rate was used as a continuous variable to control for APR-DRG, avoiding the computational burden and difficulties of entering APR-DRG itself (with >300 categories) into the multivariate model. Its use is analogous to a propensity score [15].

Thirty comorbidities were defined using methods developed by Elixhauser and colleagues [16]. This method excludes the primary reason for admission when assigning comorbidities. Additionally, patients with multiple comorbidities were classified into separate categories of 2–5 or ≥5 comorbidities. These comorbidities are also the basis for a comorbidity score based on Zhan and Miller’s [17] model, predicting mortality with the 30 comorbidities as predictor variables. Their model also adjusted for patient age, sex, minority status, private insurance payer status, and DRG-fixed effects using all discharges from the year 2000 Nationwide In-patient Sample data. The score was constructed by summing the positive, significant coefficients from the equation constructed by Zhan and Miller for each patient comorbidity. The comorbidities entering the calculation by this method were congestive heart failure, cardiac arrhythmia, peripheral vascular disease, other neurological disease, renal failure, liver disease, acquired immune deficiency syndrome/human immunodeficiency virus (AIDS/HIV), lymphoma, metastases, coagulopathy, weight loss, and fluid and electrolyte disorder.

Discharges were assigned to one of three insurance payer categories based on a combination of the primary and the secondary payers. Any discharge with private insurance as the primary or secondary payer was classified into the ‘private insurance’ group. Discharges with Medicare, medical assistance, or other governmental assistance as one of the two payers (and no private insurance) were classified into ‘governmental insurance’. Those with combinations of self-pay and other or unknown were classed as ‘other’.

Hospital characteristics. Covariates created to represent hospital characteristics were assigned to each discharge for a particular hospital: ≥8000 discharges in 2001, urban location, and percent of board-certified medical staff; percent of in-patient surgical procedures out of total facility discharges including deaths; quartiles of medical record personnel per 1000 discharges; average number of diagnostic codes used in the hospital after removing injury codes; proportion of non-specific codes out of total number of codes used in the hospital and proportion of unspecified codes out of total number of codes in the hospital (including injury codes—non-specific and unspecified as defined in the ICD-9-CM manual); teaching status categorized into none, 1–2 and 3, or more residency programs; hospital ownership (county, religious non-profit, other non-profit, and for-profit); oncology services; transplant services; the proportion of cardiac intensive care in-patient days to total in-patient days (hereafter referred to as cardiac ICU proportion); and trauma center level.

Self-reported trauma center levels were defined as follows: ‘Level 1 is a regional resource trauma center, which is capable of providing total care for every aspect of injury and plays a leadership role in trauma research and education, Level 2 is a community trauma center, which is capable of providing trauma care to all but the most severely injured patients who require specialized care, and Level 3 is a rural trauma hospital, which is capable of providing care to a large number of injury victims and can resuscitate and stabilize more severely injured patients so that they can be transported to level 1 or 2 facilities’ [18].

Nurse staffing levels were examined by the following variables: (i) registered nursing personnel per adjusted patient days, (ii) licensed practical nursing personnel per adjusted patient days, (iii) ancillary nursing personnel per adjusted patient days, (iv) total nursing personnel per adjusted patient days, and (v) percent of nursing care by registered nurses. Adjusted patient days were defined as the sum of in-patient days and outpatient visits multiplied by the ratio of outpatient revenue per outpatient visit to in-patient revenue per in-patient day for the year 2001. Nursing personnel were represented by hours per year, assuming a 40-hour workweek and 52-week year. Percent of nursing care by registered nurses was defined as the percentage of total nursing personnel adjusted patient days attributable to registered nurses.

Analyses. Logistic regression models were used to examine the association between characteristics of patient and hospital and presence or absence of a medical injury diagnosis. The likelihood of a similarity in patients at a particular hospital was accounted for by using the Huber/White/sandwich estimator of variance to obtain odds ratio confidence intervals (OR CI). Analyses were conducted using Stata 8.0 [19].

Results

In our analysis, we detected medical injury in 13.3% of the 556 899 non-newborn patient discharges from 132 Wisconsin hospitals in 2001. The 10 most frequent specific injuries are summarized in Table 1. Digestive system complications because of a procedure were most numerous, followed by post-operative infections (other).

View this table:
Table 1

Ten most common specific medical injuries

Injury descriptionICD-9-CM codeN (Rate per 1000 discharges)
Digestive system complication because of a procedure997.44441 (8.0)
Other post-operative infection998.593850 (6.9)
Cardiac complication during or resulting from a procedure997.13456 (6.2)
Adverse effect of correct drug, properly administered: antineoplastic/immunosuppressiveE933.13342 (6.0)
Other specified surgical operation—late complicationE878.83145 (5.6)
Adverse effect of correct drug, properly administered: adrenal cortical steroidsE932.02805 (5.0)
Respiratory complication resulting from a procedure997.32684 (4.8)
Mechanical complication: internal orthopedic device, implant, or graft996.42677 (4.8)
Hemorrhage complicating a procedure998.112601 (4.7)
Accidental puncture or laceration during a procedure998.22477 (4.4)
  • ICD-9-CM, International Classification of Diseases—9th Revision—Clinical Modification. See http://www.mcw.edu/medicalinjury for a complete listing of injury categories.

Patient characteristics and their bivariate (unadjusted) relationships with medical injury are summarized in Table 2, along with adjusted odds ratios (AORs). In the bivariate comparisons, older patients showed an increased risk of medical injury and younger patients showed a similar or decreased risk compared with those 25–44 years of age. Males were at significantly increased risk. Governmental insurance and private insurance had equivalent risk. Insurance by self-pay/other/unknown was associated with decreased risk and fewer procedures (mean 1.2 versus 1.5) compared with private insurance, with shorter stays than private or governmental insurance (mean 3.4, 4.3, and 5.0 days, respectively; P = 0.0001; Kruskal–Wallis test) but not with fewer procedures than governmental insurance (mean 1.2 versus 1.1). Increasing comorbidity score showed a significantly increasing risk of medical injury.

View this table:
Table 2

Rate of medical injury by patient characteristics

CharacteristicsN = 556 899Medical injury (%)Unadjusted OR (95% CI)1Adjusted OR (95% CI)2
Age in years
    <518 0667.80.8 (0.5–1.3)0.6 (0.5–0.7)
    5–1412 38012.41.3 (1.0–1.8)0.9 (0.8–1.0)
    15–2446 7247.30.7 (0.6–0.9)0.8 (0.8–0.9)
    25–44125 0649.5ReferentReferent
    45–64123 21616.21.8 (1.7–2.0)1.3 (1.2–1.3)
    65–84181 99116.61.9 (1.6–2.2)1.3 (1.2–1.4)
    ≥8549 45812.21.3 (1.1–1.6)1.0 (0.9–1.1)
Sex
    Male227 60614.9ReferentReferent
    Female329 29312.30.8 (0.8–0.8)1.0 (1.0–1.1)
Insurance
    Private398 11813.4ReferentReferent
    Governmental141 73913.71.0 (0.9–1.1)1.0 (1.0–1.1)
    Other317 0429.20.7 (0.6–0.7)0.8 (0.7–0.9)
Comorbidity score4–0.5 to 10.113.351.4 (1.3–1.4)1.2 (1.2–1.2)
  • 95% CI, 95% confidence interval; OR, odds ratio.

  • 1 Adjusted for clustering within hospitals.

  • 2 Adjusted for hospital characteristics, diagnostic coding characteristics (proportion of non-specific codes, proportion of unspecified codes, average number of codes, and quartile of medical record personnel), and clustering within hospitals and additionally for a measure of the injury rate by All Patients Refined—Diagnosis Related Group (APR-DRG) (see Methods for details); significant ORs at P < 0.05 are shown in boldface type.

  • 3 The other insurance group is made up of other, self-pay, and unknown.

  • 4 Based on Zhan and Miller’s comorbidity score [17; see Methods]; standardized to mean of 0 and standard deviation of 1.

  • 5 Overall percentage.

After adjustment, 45–64-year-olds and 65–84-year-olds were at increased risk compared with 25–44-year-olds. Those 85 years and older were at increased risk until adjustment. The age groups below age 25 generally had a reduced risk of medical injury. Females were at increased risk compared with males after adjustment. The insurance groups and comorbidity score showed patterns similar to their unadjusted results.

Table 3 summarizes hospital characteristics and both unadjusted and adjusted relationships with medical injury. Nurse staffing levels, hospital size, and urban location were not found to be associated with medical injury and were not included in the models. Several hospital characteristics were significantly related to medical injuries when unadjusted but not after adjustment for patient characteristics and other hospital characteristics: residencies, oncology services, transplant services, and in-patient surgeries. Ownership, percent of board-certified staff, trauma level, and cardiac ICU proportion all remained significantly associated with injury after adjustment. Compared with ‘other’ non-profits, patients at for-profit hospitals had a 66% reduced odds of having a medical injury, OR = 0.34 and 95% CI = 0.18–0.64. Each 10% increase in board-certified staff was associated with 14% increased odds of injury (95% CI = 7–21%). Compared with hospitals with no trauma services, those with increasing levels of trauma services showed increasing odds of a medical injury diagnosis. Each 10% increase in the proportion of cardiac intensive care in-patient days to total in-patient days was associated with 30% increased odds of a medical injury (95% CI = 5–60%).

View this table:
Table 3

Rate of medical injury in patients based on specific hospital characteristics

CharacteristicsN = 556 899Medical injury (%)Unadjusted OR (95% CI)1Adjusted OR (95% CI)2
Ownership
    County13849.20.6 (0.4–1.0)0.7 (0.4–1.3)
    Non-profit religious177 24612.30.9 (0.7–1.1)1.1 (1.0–1.2)
    Other non-profit376 84713.9ReferentReferent
    For-profit14225.30.4 (0.1–0.9)0.3 (0.2–0.6)
Residencies
    None237 48911.0ReferentReferent
    1–2153 99812.81.2 (1.0–1.4)0.9 (0.8–1.0)
    Three or more165 41217.31.7 (1.3–2.2)0.9 (0.8–1.1)
Board-certified staff (each 10% increase)46.7–100%13.331.2 (1.1–1.4)1.1 (1.1–1.2)
Oncology services
    No46 1918.0ReferentReferent
    Yes510 70813.81.9 (1.4–2.4)1.0 (0.9–1.2)
Transplant services
    No439 93511.9ReferentReferent
    Yes116 96418.91.7 (1.3–2.3)1.1 (0.9–1.2)
In-patient surgeries (each 10% increase)0.0–88.3%13.331.2 (1.1–1.2)1.0 (1.0–1.0)
Trauma level
    None161 38010.1ReferentReferent
    Rural center33 90510.21.0 (0.8–1.2)1.1 (0.9–1.3)
    Community center301 99613.91.4 (1.2–1.7)1.2 (1.0–1.3)
    Regional resource59 61821.12.4 (1.7–3.3)1.3 (1.0–1.7)
Cardiac ICU proportion (each 10% increase)0.0–6.5%13.332.6 (1.0–6.9)1.3 (1.1–1.6)
  • ICU, intensive care unit.

  • 1 Adjusted for clustering within hospitals.

  • 2 Adjusted for hospital characteristics, diagnostic coding characteristics (proportion of non-specific codes, proportion of unspecified codes, average number of codes, and quartile of medical record personnel), and clustering within hospitals and additionally for a measure of the injury rate by All Patients Refined—Diagnosis Related Group (APR-DRG) (see Methods for details).

  • 3 Overall percentage.

Table 4 summarizes the medical injury rate per 1000 discharges for the 10 most frequent APR-DRGs. Rates for both the year 2000, which was used to adjust for APR-DRG, and the year 2001 are shown. In these APR-DRGs, the injury rates for the 2 years are similar, ranging in the study data set from a low of 5.6 in vaginal delivery to a high of 197.2 for hypovolemia and electrolyte disorders.

View this table:
Table 4

Injury rate in 10 most frequent APR-DRGs

APR-DRGFrequencyInjury rate per 1000 discharges
Year 2001Year 20001Year 2001
Vaginal delivery50 4305.55.6
Simple pneumonia16 49573.775.2
Heart failure14 67981.184.2
Major joint and limb reattachment procedure of lower extremity, except for trauma13 409125.5126.4
Cesarean delivery12 44154.450.4
Uterine and adnexa procedures for carcinoma in situ and non-malignancy12 286102.1103.2
Chest pain11 09630.437.4
Hypovolemia and electrolyte disorders10 195209.2197.2
Rehabilitation9656101.3105.1
Percutaneous cardiovascular procedures without acute myocardial infarction9355114.5134.0
  • 1 Rate used to adjust for All Patients Refined—Diagnosis Related Group (APR-DRG).

Comorbidities independently and significantly increased the odds of having a medical injury (Table 5), with the exception of HIV/AIDS, pulmonary circulation disorders, and alcohol abuse. Three comorbidities had greater than four-fold odds of a medical injury compared with no comorbidities: weight loss (OR = 4.6, 95% CI = 3.5–6.0), lymphoma (OR = 4.3, 95% CI = 3.3–5.6), and renal failure (OR = 4.3, 95% CI = 3.4–5.3).

View this table:
Table 5

Rate of injury in specific comorbidities

Comorbidity1NMedical injury (%)Adjusted OR (95% CI)2
None192 5097.7Referent
Congestive heart failure151518.82.6 (2.2–3.0)
Cardiac arrhythmia452521.72.9 (2.5–3.3)
Valvular disease160813.61.6 (1.4–1.9)
Pulmonary circulation disorders10914.71.8 (1.0–3.2)
Peripheral vascular disorders158318.62.1 (1.9–2.4)
Hypertension38 36912.11.3 (1.2–1.4)
Paralysis218518.42.0 (1.7–2.4)
Other neurological disorders187215.51.8 (1.5–2.1)
Chronic pulmonary disease12 87712.01.4 (1.3–1.5)
Diabetes, uncomplicated731912.01.4 (1.3–1.5)
Diabetes, complicated142920.12.2 (1.6–3.0)
Hypothyroidism559910.01.2 (1.1–1.3)
Renal failure181731.04.3 (3.4–5.3)
Liver disease78516.11.8 (1.4–2.2)
Peptic ulcer disease excluding bleeding80812.41.4 (1.1–1.7)
HIV/AIDS7215.31.7 (0.9–3.0)
Lymphoma49131.44.3 (3.3–5.6)
Metastatic cancer263929.63.7 (3.3–4.2)
Solid tumor without metastasis658521.32.5 (2.2–2.8)
Rheumatoid arthritis/collagen vascular disease154315.01.8 (1.5–2.1)
Coagulopathy157026.03.5 (3.0–4.1)
Obesity429512.51.5 (1.4–1.7)
Weight loss43031.24.6 (3.5–6.0)
Fluid and electrolyte disorders11 55415.12.1 (1.9–2.3)
Blood loss anemia147515.72.1 (1.8–2.5)
Deficiency anemias616511.11.5 (1.4–1.8)
Alcohol abuse38689.31.2 (1.0–1.4)
Drug abuse216112.51.9 (1.5–2.3)
Psychoses313019.52.9 (2.5–3.3)
Depression701615.62.1 (1.9–2.4)
Two to four comorbidities216 16117.42.1 (1.9–2.2)
Five or more comorbidities12 83516.01.9 (1.7–2.1)
  • OR, odds ratio.

  • 1Comorbidities were defined as by Elixhauser and colleagues [16]. Patients with more than one of the listed comorbidities were classified into 2–4, or 5 or more categories.

  • 2Adjusted for age, sex, insurance payer status, hospital ownership, number of residencies, board-certified staff, oncology services, transplant services, percent of in-patient surgeries, trauma level, cardiac intensive care unit patient volume, diagnostic coding characteristics (proportion of non-specific codes, proportion of unspecified codes, average number of codes, and quartile of medical record personnel), and clustering within hospitals.

Discussion

We employed an analytic approach not previously used in examinations of medical injuries. Multivariate logistic regression models showed certain associations between covariates and medical injury found in prior studies: increased risk associated with comorbidities, reduced risk in younger people (<25 years old) and increased risk in older adults, and decreased risk at private for-profit hospitals [1,2,5,8,10,20]. We also found male sex to be associated with increased odds of a discharge diagnosis of medical injury, but after adjustment females were at increased risk. The finding of decreased risk in patients whose primary insurance was self-pay/other/unknown compared with those with private insurance may reflect a lower level of exposure to treatment, as these patients had shorter stays than those with private or governmental insurance.

Because of the size of our data set, we were able to control for many covariates. A number of covariates were included in the models to control for the range and intensity of services performed by the hospital. Hospitals that do not perform a procedure would be much less likely to have patients with an injury diagnosis specific to that procedure (although a patient who incurred such an injury elsewhere might have been admitted subsequently). Accordingly, we examined oncology services, transplant services, percent of in-patient surgeries, cardiac ICU proportion, and trauma center level. In multivariate analyses, only higher cardiac ICU proportion and trauma center level remained significant independent predictors of a medical injury diagnosis.

The significant increase in risk with a greater proportion of board-certified staff may reflect increased detection and reporting of medical injuries or a higher number of interventions that increase the opportunity for a medical injury. Some associations seen in a previous study [4] were not found in our multivariate analyses: hospital teaching status, size, or urban location. We found no significant relationships between our nurse staffing variables and medical injury [21,22]. However, our medical injury criteria were not designed to be sensitive to nurse staffing levels.

The increased risk of medical injury of women after adjustment, including controlling for APR-DRG, as compared with a reduced risk before adjustment is likely because of the large number of women hospitalized for a vaginal delivery (Table 4).

Our method introduced four variables to control for variability in hospital coding practices: (i) average number of diagnostic codes for a patient record, (ii) proportion of non-specific codes over all records for a given hospital, (iii) proportion of unspecified codes, and (iv) number of medical record personnel per 1000 discharges. Controlling for these measures of coding practices had the greatest impact on the ORs for the hospital level characteristics of ownership and residency status but relatively little impact on the patient characteristics ORs.

Definitions of medical injury vary. Studies by Brennan and colleagues [1], Wilson and colleagues [2], Thomas and colleagues [4], and Vincent and colleagues [5] defined an adverse event as ‘an injury that was caused by medical management (rather than the underlying disease) and that prolonged the hospitalization, produced a disability at the time of discharge, or both’. Additionally, cases in these studies needed to pass an initial screen consisting of 18 high-risk situations, including death [23]. We employed the definition ‘any untoward harm associated with a therapeutic or diagnostic health care intervention’ and no preliminary screen. Our definition provides a better understanding of medical injuries occurring in the community that require hospitalization but likely misses some medical injuries detectable by chart review and not by ICD-9 code alone. We applied our criteria by means of computer programming to an administrative database rather than utilizing chart reviews as was done in the Harvard, Australian, Utah/Colorado, and UK studies [1,2,4,5]. Nevertheless, the injury rate in this data set (13.3%) is within the range found in other studies.

Limitations

Although administrative data lack the clinical detail that can be obtained by chart review, they offer the advantage of being much less labor intensive. Because many covariates were examined, some significant associations with medical injury may have occurred by chance. As there are no patient identifiers, multiple hospital stays by a single patient cannot be linked. Our criteria were designed to capture medical injuries occurring in any setting for hospitalized individuals. We also could not determine whether an injury occurred before or during the current hospitalization. Thomas and colleagues [4] found that 83.8% of medical errors occurred in hospital rather than in other health care settings. In our study, the percentage of medical injuries occurring in the hospital is likely somewhat lower because of the inclusion of injuries that can occur outside the hospital.

We cannot determine to what extent medical injuries may be under-coded, but we included coding variables in the models to try to account for coding practices. To the extent that injury criteria are based on E codes, Wisconsin hospital data have very complete coding because of a state requirement for ICD-9 E codes to accompany any diagnostic code in the range 800–995 (injury and poisoning codes).

Although Elixhauser and colleagues’ method [16] attempted to exclude conditions arising during the hospitalization, it is possible that some of the comorbidities detected were actually later complications or consequences of medical injury as was found in another study using the method [24].

Conclusion

This study was an initial attempt to examine risk factors for medical injury in a comprehensive, multivariate manner. As such, it extends the bivariate and limited multivariate analyses of risk factors of medical injury and introduces a new approach to control for diagnostic coding variability. Whereas the limited clinical information available in this data set precludes the determination of causal relationships, the patterns elucidated through this multivariate analysis can be used to generate hypotheses for further testing through other methods and suggest intervention points for patient safety efforts.

References

View Abstract