Assessing the risk of intentional self-harm in montelukast users: an updated Sentinel System analysis using ICD-10 coding

BACKGROUND

Montelukast prescribing information includes a Boxed Warning issued in March 2020 regarding neuropsychiatric adverse events. A previous Sentinel System study of asthma patients from 2000 to 2015 did not demonstrate an increased risk of intentional self-harm measured using the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes, with montelukast compared to inhaled corticosteroids (ICS).

METHODS

Using a new user cohort study design, we examined intentional self-harm events in patients aged 10 years and older who were incident users of either montelukast or ICS as monotherapy, with a diagnosis of asthma, between October 1, 2015, to June 30, 2022, in the Sentinel System. We measured intentional self-harm using ICD-10-CM codes, which may have better accuracy for capturing suicide attempts than ICD-9-CM codes. We used inverse probability of treatment weighting to balance baseline covariates. We performed subgroup analyses by age group, sex, psychiatric history, and pre/post Boxed Warning era and conducted sensitivity analyses varying type of care setting of the outcome and exposure episode gaps.

RESULTS

Among 752,230 and 724,855 patients in the montelukast and ICS exposure groups respectively, we found no association between montelukast use and self-harm compared to ICS use [Hazard Ratio (95% Confidence Interval): 0.96 (0.85, 1.08)]. This finding was consistent across all subgroups, and sensitivity analyses.

CONCLUSION

Our results cannot exclude other neuropsychiatric idiosyncratic reactions to montelukast. Compared to the previous Sentinel study, this study identified about double the rate of self-harm events, suggesting a greater sensitivity of ICD-10 codes for measuring self-harm than ICD-9.

"Pathological" fractures in spinal cord injuries and disorders: Insight into International classification of diseases, ninth revision coding

OBJECTIVE

Analyses of osteoporosis-related fractures in persons with Spinal Cord Injury or Disorder (SCID) using administrative data often exclude pathological fractures (International Classification of Diseases, Ninth Revision (ICD-9) codes 733.1x). We examined how often lower extremity "pathological" fractures were secondary to osteoporosis.

DESIGN

Retrospective case-control study, fiscal years 2005-2015.

SETTING

Veterans Health Administration.

PARTICIPANTS

Veterans with SCID and an ICD-9 code for lower extremity fracture.

OUTCOME MEASURES

Clinical and SCID-related characteristics were compared in pathological and non-pathological fractures. A subset of Veterans with lower extremity fracture had data on fracture etiology from prior electronic health record (eHR) review. Of these, all with eHR-confirmed pathological fractures were considered cases. For each case, four unmatched controls with non-pathological fractures from this subset were randomly selected. Fracture etiology was compared between subsample cases and controls. We sought expert opinion from specialists who care for these fractures to understand their perspectives on what constitutes a pathological fracture and narrate our findings.

RESULTS

6,397 Veterans sustained 16,279 lower extremity fractures, including 314 (1.93%) pathological fractures in 264 Veterans. Ten of 13 (76.9%) cases of pathological fracture (76.9%) and 82.4% of non-pathological fractures were secondary to osteoporosis. Of the 19 experts surveyed, only two coded osteoporotic fractures as pathological.

CONCLUSION

Most pathological lower extremity fractures by ICD-9 codes in SCID are secondary to osteoporosis. Pathological fractures can be considered for inclusion in epidemiologic studies of osteoporosis in SCID when the risk-benefit profile for the study favors capturing all osteoporotic fractures at the expense of some misclassification.

Using ICD-9 diagnostic codes for external validation of topic models derived from primary care electronic medical record clinical text data

Background/Objectives: Unsupervised topic models are often used to facilitate improved understanding of large unstructured clinical text datasets. In this study we investigated how ICD-9 diagnostic codes, collected alongside clinical text data, could be used to establish concurrent-, convergent- and discriminant-validity of learned topic models. Design/Setting: Retrospective open cohort design. Data were collected from primary care clinics located in Toronto, Canada between 01/01/2017 through 12/31/2020. Methods: We fit a non-negative matrix factorization topic model, with K = 50 latent topics/themes, to our input document term matrix (DTM). We estimated the magnitude of association between each Boolean-valued ICD-9 diagnostic code and each continuous latent topical vector. We identified ICD-9 diagnostic codes most strongly associated with each latent topical vector; and qualitatively interpreted how these codes could be used for external validation of the learned topic model. Results: The DTM consisted of 382,666 documents and 2210 words/tokens. We correlated concurrently assigned ICD-9 diagnostic codes with learned topical vectors, and observed semantic agreement for a subset of latent constructs (e.g. conditions of the breast, disorders of the female genital tract, respiratory disease, viral infection, eye/ear/nose/throat conditions, conditions of the urinary system, and dermatological conditions, etc.). Conclusions: When fitting topic models to clinical text corpora, researchers can leverage contemporaneously collected electronic medical record data to investigate the external validity of fitted latent variable models.

An Estimation of the Incidence of Thyroiditis Among Girls in Primary Care in Spain

OBJECTIVE

As for other auto-immune processes, thyroiditis is monitored after vaccinations. The aim was to estimate the baseline incidence of thyroiditis among girls, before investigating papillomavirus vaccination as a potential risk factor.

METHODS

Observational cohort study including girls aged 9-18 years and registered between 2002-2016 in the Spanish Primary Care Database for Pharmacoepidemiological Research. Girls were followed until a thyroiditis occurred, 19 years of age, left the cohort, died, or the study ended. Anonymized records were reviewed for diagnosis confirmation (endocrine discharge letter and/or free-text comments) in a random sample. Incidence rate (IR) per 10⁵ person years (/10⁵ py) was estimated.

RESULTS

The cohort numbered 480,169 girls, of whom 641 had a record of thyroiditis: 346 autoimmune thyroiditis; 17 thyroiditis of other types; and 278 unspecified. Incidence of recorded thyroiditis increased with age, from 23.96 at age 9 years to 47.91 at age 14 years, and stabilized around 31.06-34.43 among girls aged 15-18 years. Of the 98 records reviewed, 60.2% were ‘confirmed’ cases, 32.7% ‘possible’ and 7.1% were discarded. After correction for discarded cases, IR=20.83 ‘confirmed’ cases, increasing to 32.12/10⁵ py when ‘confirmed’ plus ‘possible’ cases were included. Between 2002-2005, incidences were lower (16.28 and 20.93 cases/10⁵ py) than in the period 2007-2016 (21.17 and 33.78 cases/10⁵ py) for ‘confirmed’ and ‘confirmed’ plus ‘possible’, respectively.

CONCLUSION

Two-thirds of the recorded thyroiditis included confirmatory evidence. The incidence of thyroiditis among girls increased with age and in the later period, and remained stable among girls aged 15-18 years.

The recording and prevalence of Inflammatory bowel disease in girls' primary care medical Spanish records

PURPOSE

Inflammatory bowel disease (IBD) recording validation among girls in the Spanish Primary Care Database For Pharmacoepidemiological Research (BIFAP).

METHODS

In this observational study, girls aged 9 to 18 years registered in BIFAP between 2002 and 2016, were followed up until there was a recorded IBD diagnosis or a referral to specialist indicating IBD. Anonymized profiles were reviewed to retrieve diagnosis confirmation (a positive colonoscopy or biopsy, specialist, or physician's comments mentioning the IBD diagnosis) or discarding (negative procedure results, alternative diagnosis, or family history). "possible" IBD were profiles missing that evidence, or had suspected IBD. The prescriptions of intestinal anti-inflammatory agents, azatioprine, and mercaptopurine were collected. The prevalence of IBD was estimated after review.

RESULTS

Out of 480 634 girls, 323 had a first ever recorded IBD, of which, 37.8% (N = 122) were "confirmed" incident IBD diagnosis, 19.8% (N = 64) discarded and 38.7% (N = 125) "possible" IBD. Additionally, 12 IBD records (3.7%) referred to prevalent IBD. Prescriptions were recorded in 94.3% (confirmed), 63.2% (possible), 83.3% (prevalent), and 3.1% (discarded) IBD cases. Prevalence was 52.83 "confirmed" or 93.58/10⁵ girls when "possible" IBD were added.

CONCLUSIONS

For a third of the girls, the first recorded IBD included evidence confirming the diagnosis while most of those with missing evidence had treatment indicated for IBD. For research focused in sensitivity, an algorithm including "possible" plus "confirmed" episodes is recommended, whereas only "confirmed" to guarantee higher predictive value. Prevalence suggests that IBD is not a rare disease among girls.

Validating an algorithm for multiple myeloma based on administrative data using a SEER tumor registry and medical record review

PURPOSE

Large numbers of multiple myeloma patients can be studied in real-world clinical settings using administrative databases. The validity of these studies is contingent upon accurate case identification. Our objective was to develop and evaluate algorithms to use with administrative data to identify multiple myeloma cases.

METHODS

Patients aged ≥18 years with ≥1 International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) code for multiple myeloma (203.0x) were identified at two study sites. At site 1, several algorithms were developed and validated by comparing results to tumor registry cases. An algorithm with a reasonable positive predictive value (PPV) (0.81) and sensitivity (0.73) was selected and then validated at site 2 where results were compared with medical chart data. The algorithm required that ICD-9-CM codes 203.0x occur before and after the diagnostic procedure codes for multiple myeloma.

RESULTS

At site 1, we identified 1432 patients. The PPVs of algorithms tested ranged from 0.54 to 0.88. Sensitivities ranged from 0.30 to 0.88. At site 2, a random sample (n = 400) was selected from 3866 patients, and medical charts were reviewed by a clinician for 105 patients. Algorithm PPV was 0.86 (95% CI, 0.79-0.92).

CONCLUSIONS

We identified cases of multiple myeloma with adequate validity for claims database analyses. At least two ICD-9-CM diagnosis codes 203.0x preceding diagnostic procedure codes for multiple myeloma followed by ICD-9-CM codes within a specific time window after diagnostic procedure codes were required to achieve reasonable algorithm performance.

United States National Trends in Mortality, Length of Stay (LOS) and Associated Costs of Cognitive Impairment in HIV Population from 2005 to 2014

We evaluated national trends of in-hospital discharge rates, mortality outcomes, health care costs, length of stay in HIV patients with cognitive disorders. Neurological involvement in HIV is commonly associated with cognitive impairment termed as HIV-associated neurocognitive disorder (HAND) which includes a spectrum of neurocognitive dysfunction associated with HIV infection. Although severe and progressive neurocognitive impairment has become rare in HIV patients in the era of potent antiretroviral therapy, a majority of HIV patients have mild to moderate degree of neurocognitive impairment. Study population for this analysis was derived from the Nationwide Inpatient Sample from 2005 to 2014. Patients with ICD-9 code of HIV (042) with discharge diagnosis (Dx) listed top 1 through 5 were included in the analysis. Within this population, we identified patients with cognitive impairment using ICD-9 codes of 294 (persistent mental disorders; organic psychotic brain syndromes (chronic), 323.9 (encephalitis, myelitis, and encephalomyelitis), 331.83 (mild cognitive impairment) with Dx listed from 1 to 25. Patient variables obtained included: age, race, gender, length of stay, in-hospital mortality and insurance status. Hospital level variables included teaching status, location and region of country. SAS 9.4 software was used for data analysis. Comparisons of variables between hospitalized HIV patients with and without HAND showed significant increase in cost per hospital admissions, longer hospital stay and higher risk of mortality in patients with HAND.

A simulation study investigating power estimates in phenome-wide association studies

Background

Phenome-wide association studies (PheWAS) are a high-throughput approach to evaluate comprehensive associations between genetic variants and a wide range of phenotypic measures. PheWAS has varying sample sizes for quantitative traits, and variable numbers of cases and controls for binary traits across the many phenotypes of interest, which can affect the statistical power to detect associations. The motivation of this study is to investigate the various parameters which affect the estimation of statistical power in PheWAS, including sample size, case-control ratio, minor allele frequency, and disease penetrance.

Results

We performed a PheWAS simulation study, where we investigated variations in statistical power based on different parameters, such as overall sample size, number of cases, case-control ratio, minor allele frequency, and disease penetrance. The simulation was performed on both binary and quantitative phenotypic measures. Our simulation on binary traits suggests that the number of cases has more impact on statistical power than the case to control ratio; also, we found that a sample size of 200 cases or more maintains the statistical power to identify associations for common variants. For quantitative traits, a sample size of 1000 or more individuals performed best in the power calculations. We focused on common genetic variants (MAF > 0.01) in this study; however, in future studies, we will be extending this effort to perform similar simulations on rare variants.

Conclusions

This study provides a series of PheWAS simulation analyses that can be used to estimate statistical power for some potential scenarios. These results can be used to provide guidelines for appropriate study design for future PheWAS analyses.

Electronic supplementary material

The online version of this article (10.1186/s12859-018-2135-0) contains supplementary material, which is available to authorized users.

Association Between Type I and II Diabetes With Gallbladder Stone Disease

Objective: To assess the association of type 1 diabetes (T1DM) and type 2 diabetes (T2DM) with the subsequent development of gallbladder stone disease (GSD). Setting: Cohort Study. Participants: We identified two study cohort groups to evaluate the association of T1DM and T2DM with the development of GSD. The first group comprised a T1DM cohort of 7015 patients aged ≤ 40 years and a non-diabetes cohort randomly matched with the study cohort (4:1). The second group comprised a T2DM cohort of 51,689 patients aged ≥20 years and a non-diabetes cohort randomly matched with the study cohort (1:1). All patients were studied from 1996 to the end of 2011 or withdrawal from the National Health Insurance program to determine the incidence of GSD. Results: Compared with patients without diabetes, those with T1DM had a decreased risk of GSD [adjusted hazard ratios (aHR) = 0.48, 95% confidence interval (CI) = 0.25-0.92]. Those with T2DM had an increased risk of GSD (aHR = 1.55, 95% CI = 1.41-1.69), after adjustment for age, sex, comorbidities, and number of parity. The relative risk of GSD in the T2DM cohort was higher than that in the non-diabetes cohort in each group of age, sex, and patients with or without comorbidity. However, the relative risk of GSD in the T1DM cohort was lower than that in the non-diabetes cohort only in the age group of 20-40 years. Conclusion: Our population-based cohort study reveals a strong association between T2DM and GSD. However, an inverse relationship exists between T1DM and GSD in patients aged 20-40 years.

Comparison of ICD-9 Codes for Depression and Alcohol Misuse to Survey Instruments Suggests These Codes Should Be Used with Caution

BACKGROUND

Research suggests depression and alcohol misuse are highly prevalent among chronic hepatitis C (CHC) patients, which is of clinical concern.

AIMS

To compare ICD-9 codes for depression and alcohol misuse to validated survey instruments.

METHODS

Among CHC patients, we assessed how well electronic ICD-9 codes for depression and alcohol misuse predicted these disorders using validated instruments.

RESULTS

Of 4874 patients surveyed, 56% were male and 52% had a history of injection drug use. Based on the PHQ-8, the prevalence of depression was 30% compared to 14% based on ICD-9 codes within 12 months of survey, 37% from ICD-9 codes any time before or within 12 months after survey, and 48% from ICD-9 codes any time before or within 24 months after survey. ICD-9 codes predicting PHQ-8 depression had a sensitivity ranging from 59 to 88% and a specificity ranging from 33 to 65%. Based on the AUDIT-C, the prevalence of alcohol misuse was 21% compared to 3-23% using ICD-9 codes. The sensitivity of ICD-9 codes to predict AUDIT-C score ranged from 9 to 35% and specificity from 80 to 98%. Overall 39% of patients reported ever binge drinking, with a sensitivity of ICD-9 to predict binge drinking ranging from 7 to 33% and a specificity from 84 to 98%. More than half of patients had either an ICD-9 code for depression, a survey score indicating depression, or both (59%); more than one-third had the same patterns for alcohol misuse (36%).

CONCLUSIONS

ICD-9 codes were limited in predicting current depression and alcohol misuse, suggesting that caution should be exercised when using ICD-9 codes to assess depression or alcohol misuse among CHC patients.

International Classification of Diseases, Ninth Revision (ICD-9) Diagnosis Codes Can Identify Cerebral Venous Thrombosis in Hospitalized Adults

BACKGROUND

Cerebral venous thrombosis (CVT) is a relatively rare and understudied disease. We sought to determine the accuracy of International Classification of Diseases, Ninth Revision (ICD-9) diagnosis codes to identify CVT.

METHODS

Retrospective chart review using the electronic medical record (EMR) to identify all patients discharged with CVT following admission or emergency department visit from May 1, 2010 to May 1, 2015 at our center.

RESULTS

We identified 111 patients with an ICD-9 discharge diagnosis code of 325.0 (cerebral sinovenous thrombosis, excluding nonpyogenic cases and cases associated with pregnancy and the puerperium), 437.6 (CVT of nonpyogenic origin), or 671.5 (CVT complicating pregnancy, childbirth, or the puerperium) in any position. Of these 111 patients, 84 (75.7%) had confirmed CVT after EMR review. Searching outpatient and radiology records, we found an additional 24 patients with CVT who were not identified via query of ICD-9 discharge diagnosis codes. The ICD-9 codes 325.0, 437.6, or 671.5 in any position had a combined sensitivity of 77.8% and specificity of 92.7%; in the primary position, they had a sensitivity of 28.7% and specificity of 98.3%.

CONCLUSION

The ICD-9 codes 325.0, 437.6, and 671.5 can be used to identify CVT with acceptable sensitivity and specificity.

Classification of patients with incident non-specific low back pain: implications for research

BACKGROUND CONTEXT

Comparing research studies of low back pain is difficult because of heterogeneity. There is no consensus among researchers on inclusion criteria or the definition of an episode.

PURPOSE

This study aimed to determine pattern(s) of recurrent non-specific low back pain from data collected over 27 months.

STUDY DESIGN/SETTING

This study used retrospective cohort study using administrative claims from multiple payers. Although claims are designed for capturing costs, not clinical complexity, they are valid for describing utilization patterns, which are not affected by potential "upcoding."

PATIENT SAMPLE

The patient sample consisted of population-based, nationally generalizable sample of 65,790 adults with continuous medical and pharmaceutical commercial health insurance who received health care for incident, non-specific low back pain. Potential subjects were excluded for plausible cause of the pain, severe mental illness, or cognitive impairment.

OUTCOME MEASURES

Diagnostic and therapeutic health-care services, including medical, surgical, pharmaceutical, and complementary, received in inpatient, outpatient, and emergency settings were the outcome measures for this study.

METHODS

The methods used for this study were latent class analysis of health-care utilization over 27 months (9 quarters) following index diagnosis of non-specific low back pain occurring in January-March 2009 and an analysis sample with 60% of subjects (n=39,597) and validation sample of 40% (n=26,193).

RESULTS

Four distinct groups of patients were identified and validated. One group (53.4%) of patients recovered immediately. One third of patients (31.7%) may appear to recover over 6 months, but maintain a 37-48% likelihood of receiving care for low back pain in every subsequent quarter, implying frequent relapse. Two remaining groups of patients each maintain very high probabilities of receiving care in every quarter (65-78% and 84-90%), predominantly utilizing therapeutic services and pain medication, respectively. Probabilistic grouping relative to alternatives was very high (89.6-99.3%). Grouping was not related to demographic or clinical characteristics.

CONCLUSIONS

The four distinct sets of patient experiences have clear implications for research. Inclusion criteria should specify incident or recurrent cases. A 6-month clean period may not be sufficiently long to assess incidence. Reporting should specify the proportion recovering immediately to prevent mean recovery rates from masking between-group differences. Continuous measurement of pain or disability may be more reliable than measuring outcomes at distinct endpoints.

Quality and outcomes reporting in trauma using International Statistical Classification for Diseases, Ninth Revision codes

BACKGROUND

Use of the trauma and injury severity score (TRISS) for quality and outcomes assessment is challenged by the need for laborious collection of demographic and physiological data. We hypothesize that a novel stratification approach based on International Statistical Classification for Diseases, Ninth Revision (ICD-9) data that are readily available for trauma patients provides a more accurate and more easily obtainable alternative to TRISS with the potential for widespread use.

METHODS

Data from the ACS National Trauma Data Bank were used to train and evaluate a regularized logistic regression model for mortality and linear regression models for hospital length of stay (HLOS) and intensive care unit length of stay (ILOS) using ICD-9 diagnostic and procedural codes. Model training was performed on data from 2008 (n = 124,625) and evaluation on data from 2009 (n = 120,079). The discrimination and calibration of each model based on ICD-9 codes were compared with those of TRISS.

RESULTS

The mortality model using ICD-9 codes was comparable with that of TRISS in terms of the area under the receiver operating characteristic curve (0.922 versus 0.921, P = not significant.) and achieved better results in terms of both integrated discrimination improvement (0.106, P < 0.001) and Hosmer-Lemeshow chi-squared value (294.15 versus 2043.20). The HLOS and ILOS models using ICD-9 codes also demonstrated improvements in both R(2) (0.64 versus 0.30 for HLOS, 0.68 versus 0.34 for ILOS) and root mean-squared error (7.06 versus 8.62 for HLOS, 4.15 versus 9.54 for ILOS).

CONCLUSIONS

Use of ICD-9 codes for stratification provides a more accurate and more broadly applicable approach to quality and outcomes assessment in trauma patients than the labor-intensive gold standard of TRISS.

Validity of diagnostic codes and laboratory tests of liver dysfunction to identify acute liver failure events

PURPOSE

Identification of acute liver failure (ALF) is important for post-marketing surveillance of medications, but the validity of using ICD-9 diagnoses and laboratory data to identify these events within electronic health records is unknown. We examined positive predictive values (PPVs) of hospital ICD-9 diagnoses and laboratory tests of liver dysfunction for identifying ALF within a large, community-based integrated care organization.

METHODS

We identified Kaiser Permanente Northern California health plan members (2004-2010) with a hospital diagnosis suggesting ALF (ICD-9 570, 572.2, 572.4, 572.8, 573.3, 573.8, or V42.7) plus an inpatient international normalized ratio ≥1.5 (off warfarin) and total bilirubin ≥5.0 mg/dL. Hospital records were reviewed by hepatologists to adjudicate ALF events. PPVs for confirmed outcomes were determined for individual ICD-9 diagnoses, diagnoses plus prescriptions for hepatic encephalopathy treatment, and combinations of diagnoses in the setting of coagulopathy and hyperbilirubinemia.

RESULTS

Among 669 members with no pre-existing liver disease, chart review confirmed ALF in 62 (9%). Despite the presence of co-existing coagulopathy and hyperbilirubinemia, individual ICD-9 diagnoses had low PPVs (range, 5-15%); requiring prescriptions for encephalopathy treatment did not increase PPVs of these diagnoses (range, 2-23%). Hospital diagnoses of other liver disorders (ICD-9 573.8) plus hepatic coma (ICD-9 572.2) had high PPV (67%; 95%CI, 9-99%) but only identified two (3%) ALF events.

CONCLUSIONS

Algorithms comprising relevant hospital diagnoses, laboratory evidence of liver dysfunction, and prescriptions for hepatic encephalopathy treatment had low PPVs for confirmed ALF events. Studies of ALF will need to rely on medical records to confirm this outcome.

Validity of diagnostic codes to identify cases of severe acute liver injury in the US Food and Drug Administration's Mini-Sentinel Distributed Database

PURPOSE

The validity of International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes to identify diagnoses of severe acute liver injury (SALI) is not well known. We examined the positive predictive values (PPVs) of hospital ICD-9-CM diagnoses in identifying SALI among health plan members in the Mini-Sentinel Distributed Database (MSDD) for patients without liver/biliary disease and for those with chronic liver disease (CLD).

METHODS

We selected random samples of members (149 without liver/biliary disease; 75 with CLD) with a principal hospital diagnosis suggestive of SALI (ICD-9-CM 570, 572.2, 572.4, 572.8, 573.3, 573.8, or V42.7) in the MSDD (2009-2010). Medical records were reviewed by hepatologists to confirm SALI events. PPVs of codes and code combinations for confirmed SALI were determined by CLD status.

RESULTS

Among 105 members with available records and no liver/biliary disease, SALI was confirmed in 26 (PPV, 24.7%; 95%CI, 16.9-34.1%). Combined hospital diagnoses of acute hepatic necrosis (570) and liver disease sequelae (572.8) had high PPV (100%; 95%CI, 59.0-100%) and identified 7/26 (26.9%) events. Among 46 CLD members with available records, SALI was confirmed in 19 (PPV, 41.3%; 95%CI, 27.0-56.8%). Acute hepatic necrosis (570) or hepatorenal syndrome (572.4) plus any other SALI code had a PPV of 83.3% (95%CI, 51.6-97.9%) and identified 10/19 (52.6%) events.

CONCLUSIONS

Most individual hospital ICD-9-CM diagnoses had low PPV for confirmed SALI events. Select code combinations had high PPV but did not capture all events.