Effective Identification of Similar Patients Through Sequential Matching over ICD Code Embedding.

Evidence-based medicine often involves the identification of patients with similar conditions, which are often captured in ICD (International Classification of Diseases (World Health Organization 2013)) code sequences. With no satisfying prior solutions for matching ICD-10 code sequences, this paper presents a method which effectively captures the clinical similarity among routine patients who have multiple comorbidities and complex care needs. Our method leverages the recent progress in representation learning of individual ICD-10 codes, and it explicitly uses the sequential order of codes for matching. Empirical evaluation on a state-wide cancer data collection shows that our proposed method achieves significantly higher matching performance compared with state-of-the-art methods ignoring the sequential order. Our method better identifies similar patients in a number of clinical outcomes including readmission and mortality outlook. Although this paper focuses on ICD-10 diagnosis code sequences, our method can be adapted to work with other codified sequence data.

Proportion of patients with cancer among high-cost Medicare beneficiaries: Who they are and what drives their spending.

BACKGROUND: A small proportion of patients account for the majority of health care spending. Of this group, little is known about what proportion have a cancer diagnosis and how their spending pattern compares to those without cancer. METHODS: Using national Medicare data of enrollees 65 or older, we identified patients in the top decile of spending in 2014 and designated them as high-cost. We used ICD-9 codes to identify patients with a cancer diagnosis and examined cancer prevalence among both high-cost and non-high-cost patients. We examined patterns of spending for high-cost patients with and without cancer. RESULTS: While 14.8% of all Medicare beneficiaries have a cancer diagnosis, we found that the prevalence of a cancer diagnosis was much higher among high-cost patients (32.5% versus 12.9% of non-high-cost patients). Thus, having a cancer diagnosis was associated with a 3.1 times greater odds of being high-cost, even after accounting for age (odds ratio 3.09, 95% CI 3.07-3.11; P < 0.001). High-cost patients with cancer had higher total annual spending than high-cost patients without cancer ($66,685 vs. $59,427; p < 0.0001); costs among high-cost cancer patients were driven by greater use of outpatient treatments (19.2% of total spending vs. 13.6% among non-cancer high-cost patients, p < 0.0001) and more prescription drugs (11.9% vs. 9.9%; p < 0.0001). CONCLUSIONS: There is a high prevalence of cancer diagnoses among high-cost Medicare patients. IMPLICATIONS: Programs that target high-cost patients may need to customize interventions based on whether the patient has a cancer diagnosis.

Mental disorder sick leave in Sweden: A population study.

BACKGROUND: The inability to perform productive work due to mental disorders is a growing concern in advanced societies. OBJECTIVE: To investigate medically certified mental disorder and all-cause sick leave in a working population using demographic, socioeconomic and occupational predictors. METHODS: The study population was the entire Swedish work force aged 16-64 years in December 31st 2011. The outcome was sick leave exceeding 14 days in 2012 with adjustment for 13 confounders. RESULTS: The risk of sick leave with a mental disorder is higher among women compared to men, among persons aged 30-39 and among parents in families with underage children. Employees in welfare service occupations within health care, education and social services have an elevated risk of mental disorder sick leave and constitute a large proportion of the workforce. CONCLUSION: The results support the need for improving early detection and prevention of mental disorders in the workforce. Improvements in psychosocial work environments are essential, where the higher risk in female dominated welfare occupations particularly, have repercussions on the quality of the welfare services provided for vulnerable groups in society. Better work-life balance in families with younger children could also mitigate the effects of a high total workload in that particular phase of life.

Exploring the relationship between nonalcoholic fatty liver disease and pancreatic cancer by computed tomographic survey.

Pancreatic cancer is difficult to diagnose in an early stage, and has the highest mortality of all types of cancer. Obesity, high body mass index, and increased abdominal girth are established risk factors. Some studies have postulated that there is a correlation between organ steatosis and pancreatic cancer. This study aims to explore whether nonalcoholic fatty liver disease (NAFLD) is a risk factor and a prognostic factor for pancreatic cancer. The study enrolled 557 patients (143 with and 414 without pancreatic cancer) who were diagnosed between January 2009 and December 2013. We reviewed the abdominal computed tomographic scans of the patients to confirm the diagnosis of NAFLD. Clinical parameters, laboratory data, and personal information were analyzed. NAFLD is an independent risk factor for pancreatic cancer according to adjusted multivariate logistic regression analysis (OR 2.63, 95% CI 1.24-5.58, p = 0.011). The Kaplan-Meier survival curve reveals that patients without NAFLD have longer survival than patients with NAFLD (p = 0.005, log-rank test). NAFLD is positively correlated with pancreatic cancer, a result suggesting that NAFLD may increase the incidence and risk of pancreatic cancer. Patients with pancreatic cancer and NAFLD have poorer overall survival than patients without NAFLD, perhaps, because dysregulated cytokine status leads to progression of pancreatic cancer. NAFLD may be a prognostic factor for pancreatic cancer.

The Value of Electronically Extracted Data for Auditing Outpatient Antimicrobial Prescribing.

OBJECTIVE The optimal approach to auditing outpatient antimicrobial prescribing has not been established. We assessed how different types of electronic data-including prescriptions, patient-visits, and International Classification of Disease, Tenth Revision (ICD-10) codes-could inform automated antimicrobial audits. DESIGN Outpatient visits during 2016 were retrospectively reviewed, including chart abstraction, if an antimicrobial was prescribed (cohort 1) or if the visit was associated with an infection-related ICD-10 code (cohort 2). Findings from cohorts 1 and 2 were compared. SETTING Primary care clinics and the emergency department (ED) at the Iowa City Veterans Affairs Medical Center. RESULTS In cohort 1, we reviewed 2,353 antimicrobial prescriptions across 52 providers. ICD-10 codes had limited sensitivity and positive predictive value (PPV) for validated cases of cystitis and pneumonia (sensitivity, 65.8%, 56.3%, respectively; PPV, 74.4%, 52.5%, respectively). The volume-adjusted antimicrobial prescribing rate was 13.6 per 100 ED visits and 7.5 per 100 primary care visits. In cohort 2, antimicrobials were not indicated in 474 of 851 visits (55.7%). The antimicrobial overtreatment rate was 48.8% for the ED and 59.7% for primary care. At the level of the individual prescriber, there was a positive correlation between a provider's volume-adjusted antimicrobial prescribing rate and the individualized rates of overtreatment in both the ED (r=0.72; P<.01) and the primary care setting (r=0.82; P=0.03). CONCLUSIONS In this single-center study, ICD-10 codes had limited sensitivity and PPV for 2 infections that typically require antimicrobials. Electronically extracted data on a provider's rate of volume-adjusted antimicrobial prescribing correlated with the frequency at which unnecessary antimicrobials were prescribed, but this may have been driven by outlier prescribers. Infect Control Hosp Epidemiol 2018;39:64-70.

Extracorporeal membrane oxygenation for life-threatening asthma refractory to mechanical ventilation: analysis of the Extracorporeal Life Support Organization registry.

BACKGROUND: The use of extracorporeal membrane oxygenation (ECMO) in cases of near-fatal asthma (NFA) has increased, but the benefits and potential complications of this therapy have yet to be fully investigated. METHODS: Cases were extracted from the Extracorporeal Life Support Organization Registry between March 1992 and March 2016. All patients with a diagnosis of asthma (according to the International Classification of Diseases 9th edition), who also received ECMO, were extracted. Exclusion criteria included patients who underwent multiple courses of ECMO; those who received ECMO for cardiopulmonary resuscitation or cardiac dysfunction; and those with another primary diagnosis, such as sepsis. We analyzed survival to hospital discharge, complications, and clinical factors associated with in-hospital mortality, in patients with severe life-threatening NFA requiring ECMO support. RESULTS: In total 272 patients were included. The mean time spent on ECMO was 176.4 hours. Ventilator settings, including rate, fraction of inspired oxygen (FiO2), peak inspiratory pressure (PIP), and mean airway pressure, significantly improved after ECMO initiation (rate (breaths/min), 19.0 vs. 11.3, p < 0.001; FiO2 (%), 81.2 vs. 48.8, p < 0.001; PIP (cmH2O), 38.2 vs. 25.0, p < 0.001; mean airway pressure (cmH2O): 21.4 vs. 14.2, p < 0.001). In particular, driving pressure was significantly decreased after ECMO support (29.5 vs. 16.8 cmH2O, p < 0.001). The weaning success rate was 86.7%, and the rate of survival to hospital discharge was 83.5%. The total complication rate was 65.1%, with hemorrhagic complications being the most common (28.3%). Other complications included renal (26.8%), cardiovascular (26.1%), mechanical (24.6%), metabolic (22.4%), infection (16.5%), neurologic (4.8%), and limb ischemia (2.6%). Of the hemorrhagic complications, cannulation site hemorrhage was the most common (13.6%). Using multivariate logistic regression analysis, it was found that hemorrhage was associated with increased in-hospital mortality (odds ratio, 2.97; 95% confidence interval, 1.07-8.24; p = 0.036). Hemorrhage-induced death occurred in four patients (1.5%). The most common reason for death was organ failure (37.8%). CONCLUSIONS: ECMO can provide adequate gas exchange and prevent lung injury induced by mechanical ventilation, and may be an effective bridging strategy to avoid aggressive ventilation in refractory NFA. However, careful management is required to avoid complications.

[Validity of the ICD-10 Symptom Rating (ISR) in a Non-Clinical Sample]. / Faktorielle Validität und Konstruktvalidität des ICD-10-Symptom-Ratings (ISR) in einer klinisch weitgehend unauffälligen Stichprobe.

Introduction The ICD-10 symptom rating (ISR) is a self-rating instrument that is based on ICD-10 syndromes. Can previous findings concerning its validity be replicated in a non-clinical sample? Material & Methods N=428 study participants - mainly students - completed the ISR and the SCL-90-R in an online survey. Results The assumed factorial structure was replicated with a good model fit. The correlations between the content-related scales of the two instruments ranged from rmin=0.60 to rmax=0.85. Study participants indicated that they did not find completing the ISR stressful. Discussion and Conclusions These results indicate good validity and applicability of the ISR.

The impact of the 2006 Massachusetts health care reform law on spine surgery patient payer-mix status and age.

OBJECTIVE Several similarities exist between the Massachusetts health care reform law of 2006 and the Affordable Care Act (ACA). The authors' prior neurosurgical research showed a decrease in uninsured surgeries without a significant change in surgical volume after the Massachusetts reform. An analysis of the payer-mix status and the age of spine surgery patients, before and after the policy, should provide insight into the future impact of the ACA on spine surgery in the US. METHODS Using the Massachusetts State Inpatient Database and spine ICD-9-CM procedure codes, the authors obtained demographic information on patients undergoing spine surgery between 2001 and 2012. Payer-mix status was assigned as Medicare, Medicaid, private insurance, uninsured, or other, which included government-funded programs and workers' compensation. A comparison of the payer-mix status and patient age, both before and after the policy, was performed. The New York State data were used as a control. RESULTS The authors analyzed 81,821 spine surgeries performed in Massachusetts and 248,757 in New York. After 2008, there was a decrease in uninsured and private insurance spine surgeries, with a subsequent increase in the Medicare and "other" categories for Massachusetts. Medicaid case numbers did not change. This correlated to an increase in surgeries performed in the age group of patients 65-84 years old, with a decrease in surgeries for those 18-44 years old. New York showed an increase in all insurance categories and all adult age groups. CONCLUSIONS After the Massachusetts reform, spine surgery decreased in private insurance and uninsured categories, with the majority of these surgeries transitioning to Medicare. Moreover, individuals who were younger than 65 years did not show an increase in spine surgeries, despite having greater access to health insurance. In a health care system that requires insurance, the decrease in private insurance is primarily due to an increasing elderly population. The Massachusetts model continues to show that this type of policy is not causing extreme shifts in the payer mix, and suggests that spine surgery will continue to thrive in the current US health care system.

Trends in pneumococcal meningitis hospitalizations following the introduction of the 13-valent pneumococcal conjugate vaccine in the United States.

BACKGROUND: The 13-valent pneumococcal conjugate vaccine (PCV13) was introduced in 2010 in the U.S. and its impact on pneumococcal meningitis (PM) is unknown. We assessed the impact of PCV13 on PM hospitalization rates 4years after the vaccine was introduced. METHODS: This was a retrospective analysis of the National Inpatient Sample from 2008-2014. Patients with an ICD-9-CM code for PM (320.1) were identified and rates calculated using US Census data as the denominator. Data weights were used to derive national estimates. We examined three time periods: 2008-2009 (late post-PCV7), 2010 (transition year), and 2011-2014 (post-PCV13). RESULTS: During the study period, there were 10,493 hospitalizations due to PM in the U.S. Overall, PM incidence decreased from 0.62 to 0.38 cases per 100,000 over this time (39% decrease; P<0.01). Among children <2years, the average annualized PM rate decreased by 45% from 2.19 to 1.20 per 100,000 (P=0.10). Annual PM rates decreased in those aged 18-39years (0.25-0.15 cases per 100,000; P=0.02) and 40-64years (0.95-0.54 cases per 100,000; P=0.03). A total of 1016 deaths were due to PM, and the case fatality rate was variable over the study period (8.3%-11.2%; P=0.96). CONCLUSION: Following the introduction of PCV13, hospitalization rates for PM decreased significantly with no subsequent improvements in case-fatality rate.

Trends in Opioid-related Inpatient Stays Shifted After the US Transitioned to ICD-10-CM Diagnosis Coding in 2015.

BACKGROUND: Trend analyses of opioid-related inpatient stays depend on the availability of comparable data over time. In October 2015, the US transitioned diagnosis coding from International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) to ICD-10-CM, increasing from ∼14,000 to 68,000 codes. This study examines how trend analyses of inpatient stays involving opioid diagnoses were affected by the transition to ICD-10-CM. SUBJECTS: Data are from Healthcare Cost and Utilization Project State Inpatient Databases for 14 states in 2015-2016, representing 26% of acute care inpatient discharges in the US. STUDY DESIGN: We examined changes in the number of opioid-related stays before, during, and after the transition to ICD-10-CM using quarterly ICD-9-CM data from 2015 and quarterly ICD-10-CM data from the fourth quarter of 2015 and the first 3 quarters of 2016. RESULTS: Overall, stays involving any opioid-related diagnosis increased by 14.1% during the ICD transition-which was preceded by a much lower 5.0% average quarterly increase before the transition and followed by a 3.5% average increase after the transition. In stratified analysis, stays involving adverse effects of opioids in therapeutic use showed the largest increase (63.2%) during the transition, whereas stays involving abuse and poisoning diagnoses decreased by 21.1% and 12.4%, respectively. CONCLUSIONS: The sharp increase in opioid-related stays overall during the transition to ICD-10-CM may indicate that the new classification system is capturing stays that were missed by ICD-9-CM data. Estimates of stays involving other diagnoses may also be affected, and analysts should assess potential discontinuities in trends across the ICD transition.

International Classification of Disease Coding of Exertional Heat Illness in U.S. Army Soldiers.

INTRODUCTION: The severity of exertional heat illnesses (EHI) ranges from relatively minor heat exhaustion to potentially life-threatening heat stroke. Epidemiological surveillance of the types of and trends in EHI incidence depends on application of the appropriate International Classification of Disease, 9th Revision (ICD-9) diagnostic code. However, data examining whether the appropriate EHI ICD-9 code is selected are lacking. The purpose of this study was to determine whether the appropriate ICD-9 code is selected in a cohort of EHI casualties. MATERIALS AND METHODS: Chart reviews of 290 EHI casualties that occurred in U.S. Army soldiers from 2009 to 2012 were conducted. The ICD-9 diagnostic code was extracted, as were the initial and peak values for aspartate transaminase, alanine transaminase, creatine kinase, and creatinine. Diagnostic criteria for heat injury and heat stroke include evidence of organ and/or tissue damage; 2 out of 3 of the following must have been met to be considered heat injury (ICD-9 code 992.8) or heat stroke (ICD-9 code 992.0): aspartate transaminase/ alanine transaminase fold increase >3, creatine kinase fold increase >5, and/or creatinine ≥1.5mg/dL. Contingency tables were constructed from which sensitivity, specificity, and positive and negative predictive value were calculated. RESULTS: The 290 cases in this cohort represent ∼29% of all EHI at Fort Benning and ∼6% of all EHI Army-wide during the study period. There were 80 cases that met the laboratory diagnostic criteria for heat injury/stroke, however of those, 28 cases were diagnosed as an EHI other than heat injury/stroke (sensitivity = 0.65). 210 cases did not meet the laboratory diagnostic criteria, but 66 of those were incorrectly diagnosed as heat injury or heat stroke (specificity = 0.69). Positive and negative predictive values were 0.44 and 0.84, respectively. In total, the incorrect ICD-9 code was applied to 94 of 290 total cases. CONCLUSIONS: Our data suggest that caution is warranted when examining epidemiological surveillance data on EHI severity, as there was disagreement between the laboratory data and the selected ICD-9 code in ∼1/3 of all cases in this cohort. Of note is the lack of an ICD-9 or -10 code for heat injury; we recommend coding for heat exhaustion as the primary diagnosis and additional codes to capture the accompanying muscle, tissue, and/or organ damage.

Bayesian analysis of genetic association across tree-structured routine healthcare data in the UK Biobank.

Genetic discovery from the multitude of phenotypes extractable from routine healthcare data can transform understanding of the human phenome and accelerate progress toward precision medicine. However, a critical question when analyzing high-dimensional and heterogeneous data is how best to interrogate increasingly specific subphenotypes while retaining statistical power to detect genetic associations. Here we develop and employ a new Bayesian analysis framework that exploits the hierarchical structure of diagnosis classifications to analyze genetic variants against UK Biobank disease phenotypes derived from self-reporting and hospital episode statistics. Our method displays a more than 20% increase in power to detect genetic effects over other approaches and identifies new associations between classical human leukocyte antigen (HLA) alleles and common immune-mediated diseases (IMDs). By applying the approach to genetic risk scores (GRSs), we show the extent of genetic sharing among IMDs and expose differences in disease perception or diagnosis with potential clinical implications.

ICD Social Codes: An Underutilized Resource for Tracking Social Needs.

BACKGROUND: Social determinants of health (SDH) data collected in health care settings could have important applications for clinical decision-making, population health strategies, and the design of performance-based incentives and penalties. One source for cataloging SDH data is the International Statistical Classification of Diseases and Related Health Problems (ICD). OBJECTIVE: To explore how SDH are captured with ICD Ninth revision SDH V codes in a national inpatient discharge database. MATERIALS AND METHODS: Data come from the 2013 Healthcare Cost and Utilization Project (HCUP) National Inpatient Sample, a national stratified sample of discharges from 4363 hospitals from 44 US states. We estimate the rate of ICD-9 SDH V code utilization overall and by patient demographics and payer categories. We additionally estimate the rate of SDH V code utilization for: (a) the 5 most common reasons for hospitalization; and (b) the 5 conditions with the highest rates of SDH V code utilization. RESULTS: Fewer than 2% of overall discharges in the National Inpatient Sample were assigned an SDH V code. There were statistically significant differences in the rate of overall SDH V code utilization by age categories, race/ethnicity, sex, and payer (all P<0.001). Nevertheless, SDH V codes were assigned to <7% of discharges in any demographic or payer subgroup. SDH V code utilization was highest for major diagnostic categories related to mental health and alcohol/substance use-related discharges. CONCLUSIONS: SDH V codes are infrequently utilized in inpatient settings for discharges other than those related to mental health and alcohol/substance use. Utilization incentives will likely need to be developed to realize the potential benefits of cataloging SDH information.

Use of Emergency Department Data to Monitor and Respond to an Increase in Opioid Overdoses in New Hampshire, 2011-2015.

OBJECTIVES: Opioid-related overdoses and deaths in New Hampshire have increased substantially in recent years, similar to increases observed across the United States. We queried emergency department (ED) data in New Hampshire to monitor opioid-related ED encounters as part of the public health response to this health problem. METHODS: We obtained data on opioid-related ED encounters for the period January 1, 2011, through December 31, 2015, from New Hampshire's syndromic surveillance ED data system by querying for (1) chief complaint text related to the words "fentanyl," "heroin," "opiate," and "opioid" and (2) opioid-related International Classification of Diseases ( ICD) codes. We then analyzed the data to calculate frequencies of opioid-related ED encounters by age, sex, residence, chief complaint text values, and ICD codes. RESULTS: Opioid-related ED encounters increased by 70% during the study period, from 3300 in 2011 to 5603 in 2015; the largest increases occurred in adults aged 18-29 and in males. Of 20 994 total opioid-related ED visits, we identified 18 554 (88%) using ICD code alone, 690 (3%) using chief complaint text alone, and 1750 (8%) using both chief complaint text and ICD code. For those encounters identified by ICD code only, the corresponding chief complaint text included varied and nonspecific words, with the most common being "pain" (n = 3335, 18%), "overdose" (n = 1555, 8%), "suicidal" (n = 816, 4%), "drug" (n = 803, 4%), and "detox" (n = 750, 4%). Heroin-specific encounters increased by 827%, from 4% of opioid-related encounters in 2011 to 24% of encounters in 2015. CONCLUSIONS: Opioid-related ED encounters in New Hampshire increased substantially from 2011 to 2015. Data from New Hampshire's ED syndromic surveillance system provided timely situational awareness to public health partners to support the overall response to the opioid epidemic.

Evaluation of the Components of the North Carolina Syndromic Surveillance System Heat Syndrome Case Definition.

OBJECTIVES: To improve heat-related illness surveillance, we evaluated and refined North Carolina's heat syndrome case definition. METHODS: We analyzed North Carolina emergency department (ED) visits during 2012-2014. We evaluated the current heat syndrome case definition (ie, keywords in chief complaint/triage notes or International Classification of Diseases, Ninth Revision, Clinical Modification [ ICD-9-CM] codes) and additional heat-related inclusion and exclusion keywords. We calculated the positive predictive value and sensitivity of keyword-identified ED visits and manually reviewed ED visits to identify true positives and false positives. RESULTS: The current heat syndrome case definition identified 8928 ED visits; additional inclusion keywords identified another 598 ED visits. Of 4006 keyword-identified ED visits, 3216 (80.3%) were captured by 4 phrases: "heat ex" (n = 1674, 41.8%), "overheat" (n = 646, 16.1%), "too hot" (n = 594, 14.8%), and "heatstroke" (n = 302, 7.5%). Among the 267 ED visits identified by keyword only, a burn diagnosis or the following keywords resulted in a false-positive rate >95%: "burn," "grease," "liquid," "oil," "radiator," "antifreeze," "hot tub," "hot spring," and "sauna." After applying the revised inclusion and exclusion criteria, we identified 9132 heat-related ED visits: 2157 by keyword only, 5493 by ICD-9-CM code only, and 1482 by both (sensitivity = 27.0%, positive predictive value = 40.7%). Cases identified by keywords were strongly correlated with cases identified by ICD-9-CM codes (rho = .94, P < .001). CONCLUSIONS: Revising the heat syndrome case definition through the use of additional inclusion and exclusion criteria substantially improved the accuracy of the surveillance system. Other jurisdictions may benefit from refining their heat syndrome case definition.

Using Chief Complaint in Addition to Diagnosis Codes to Identify Falls in the Emergency Department.

OBJECTIVES: To compare incidence of falls in an emergency department (ED) cohort using a traditional International Classification of Diseases, Ninth Revision (ICD-9) code-based scheme and an expanded definition that included chief complaint information and to examine the clinical characteristics of visits "missed" in the ICD-9-based scheme. DESIGN: Retrospective electronic record review. SETTING: Academic medical center ED. PARTICIPANTS: Individuals aged 65 and older seen in the ED between January 1, 2013, and September 30, 2015. MEASUREMENTS: Two fall definitions were applied (individually and together) to the cohort: an ICD-9-based definition and a chief complaint definition. Admission rates and 30-day mortality (per encounter) were measured for each definition. RESULTS: Twenty-three thousand eight hundred eighty older adult visits occurred during the study period. Using the most-inclusive definition (ICD-9 code or chief complaint indicating a fall), 4,363 visits (18%) were fall related. Of these visits, 3,506 (80%) met the ICD-9 definition for a fall-related visit, and 2,664 (61%) met the chief complaint definition. Of visits meeting the chief complaint definition, 857 (19.6%) were missed when applying the ICD-9 definition alone. Encounters missed using the ICD-9 definition were less likely to lead to an admission (42.9%, 95% confidence interval (CI) = 39.7-46.3%) than those identified (54.4%, 95% CI = 52.7-56.0%). CONCLUSION: Identifying individuals in the ED who have fallen based on diagnosis codes underestimates the true burden of falls. Individuals missed according to the code-based definition were less likely to have been admitted than those who were captured. These findings call attention to the value of using chief complaint information to identify individuals who have fallen in the ED-for research, clinical care, or policy reasons.

Validation of a Syndromic Case Definition for Detecting Emergency Department Visits Potentially Related to Marijuana.

OBJECTIVES: Reliable methods are needed to monitor the public health impact of changing laws and perceptions about marijuana. Structured and free-text emergency department (ED) visit data offer an opportunity to monitor the impact of these changes in near-real time. Our objectives were to (1) generate and validate a syndromic case definition for ED visits potentially related to marijuana and (2) describe a method for doing so that was less resource intensive than traditional methods. METHODS: We developed a syndromic case definition for ED visits potentially related to marijuana, applied it to BioSense 2.0 data from 15 hospitals in the Denver, Colorado, metropolitan area for the period September through October 2015, and manually reviewed each case to determine true positives and false positives. We used the number of visits identified by and the positive predictive value (PPV) for each search term and field to refine the definition for the second round of validation on data from February through March 2016. RESULTS: Of 126 646 ED visits during the first period, terms in 524 ED visit records matched ≥1 search term in the initial case definition (PPV, 92.7%). Of 140 932 ED visits during the second period, terms in 698 ED visit records matched ≥1 search term in the revised case definition (PPV, 95.7%). After another revision, the final case definition contained 6 keywords for marijuana or derivatives and 5 diagnosis codes for cannabis use, abuse, dependence, poisoning, and lung disease. CONCLUSIONS: Our syndromic case definition and validation method for ED visits potentially related to marijuana could be used by other public health jurisdictions to monitor local trends and for other emerging concerns.

Limitations of pulmonary embolism ICD-10 codes in emergency department administrative data: let the buyer beware.

BACKGROUND: Administrative data is a useful tool for research and quality improvement; however, validity of research findings based on these data depends on their reliability. Diagnoses assigned by physicians are subsequently converted by nosologists to ICD-10 codes (International Statistical Classification of Diseases and Related Health Problems, 10th Revision). Several groups have reported ICD-9 coding errors in inpatient data that have implications for research, quality improvement, and policymaking, but few have assessed ICD-10 code validity in ambulatory care databases. Our objective was to evaluate pulmonary embolism (PE) ICD-10 code accuracy in our large, integrated hospital system, and the validity of using these codes for operational and health services research using ED ambulatory care databases. METHODS: Ambulatory care data for patients (age ≥ 18 years) with a PE ICD-10 code (I26.0 and I26.9) were obtained from the records of four urban EDs between July 2013 to January 2015. PE diagnoses were confirmed by reviewing medical records and imaging reports. In cases where chart diagnosis and ICD-10 code were discrepant, chart review was considered correct. Physicians' written discharge diagnoses were also searched using 'pulmonary embolism' and 'PE', and patients who were diagnosed with PE but not coded as PE were identified. Coding discrepancies were quantified and described. RESULTS: One thousand, four hundred and fifty-three ED patients had a PE ICD-10 code. Of these, 257 (17.7%) were false positive, with an incorrectly assigned PE code. Among the 257 false positives, 193 cases had ambiguous ED diagnoses such as 'rule out PE' or 'query PE', while 64 cases should have had non-PE codes. An additional 117 patients (8.90%) with a PE discharge diagnosis were incorrectly assigned a non-PE ICD-10 code (false negative group). The sensitivity of PE ICD-10 codes in this dataset was 91.1% (95%CI, 89.4-92.6) with a specificity of 99.9% (95%CI, 99.9-99.9). The positive and negative predictive values were 82.3% (95%CI, 80.3-84.2) and 99.9% (95%CI, 99.9-99.9), respectively. CONCLUSIONS: Ambulatory care data, like inpatient data, are subject to coding errors. This confirms the importance of ICD-10 code validation prior to use. The largest proportion of coding errors arises from ambiguous physician documentation; therefore, physicians and data custodians must ensure that quality improvement processes are in place to promote ICD-10 coding accuracy.

CodeMapper: semiautomatic coding of case definitions. A contribution from the ADVANCE project.

BACKGROUND: Assessment of drug and vaccine effects by combining information from different healthcare databases in the European Union requires extensive efforts in the harmonization of codes as different vocabularies are being used across countries. In this paper, we present a web application called CodeMapper, which assists in the mapping of case definitions to codes from different vocabularies, while keeping a transparent record of the complete mapping process. METHODS: CodeMapper builds upon coding vocabularies contained in the Metathesaurus of the Unified Medical Language System. The mapping approach consists of three phases. First, medical concepts are automatically identified in a free-text case definition. Second, the user revises the set of medical concepts by adding or removing concepts, or expanding them to related concepts that are more general or more specific. Finally, the selected concepts are projected to codes from the targeted coding vocabularies. We evaluated the application by comparing codes that were automatically generated from case definitions by applying CodeMapper's concept identification and successive concept expansion, with reference codes that were manually created in a previous epidemiological study. RESULTS: Automated concept identification alone had a sensitivity of 0.246 and positive predictive value (PPV) of 0.420 for reproducing the reference codes. Three successive steps of concept expansion increased sensitivity to 0.953 and PPV to 0.616. CONCLUSIONS: Automatic concept identification in the case definition alone was insufficient to reproduce the reference codes, but CodeMapper's operations for concept expansion provide an effective, efficient, and transparent way for reproducing the reference codes.