Development and utility assessment of a machine learning bloodstream infection classifier in pediatric patients receiving cancer treatments.

BACKGROUND: Objectives were to build a machine learning algorithm to identify bloodstream infection (BSI) among pediatric patients with cancer and hematopoietic stem cell transplantation (HSCT) recipients, and to compare this approach with presence of neutropenia to identify BSI. METHODS: We included patients 0-18 years of age at cancer diagnosis or HSCT between January 2009 and November 2018. Eligible blood cultures were those with no previous blood culture (regardless of result) within 7 days. The primary outcome was BSI. Four machine learning algorithms were used: elastic net, support vector machine and two implementations of gradient boosting machine (GBM and XGBoost). Model training and evaluation were performed using temporally disjoint training (60%), validation (20%) and test (20%) sets. The best model was compared to neutropenia alone in the test set. RESULTS: Of 11,183 eligible blood cultures, 624 (5.6%) were positive. The best model in the validation set was GBM, which achieved an area-under-the-receiver-operator-curve (AUROC) of 0.74 in the test set. Among the 2236 in the test set, the number of false positives and specificity of GBM vs. neutropenia were 508 vs. 592 and 0.76 vs. 0.72 respectively. Among 139 test set BSIs, six (4.3%) non-neutropenic patients were identified by GBM. All received antibiotics prior to culture result availability. CONCLUSIONS: We developed a machine learning algorithm to classify BSI. GBM achieved an AUROC of 0.74 and identified 4.3% additional true cases in the test set. The machine learning algorithm did not perform substantially better than using presence of neutropenia alone to predict BSI.

Classification of bloodstream infections in patients recently discharged from acute-care facilities: Hospital acquired or healthcare-associated community onset?

Laboratory-identified bloodstream infections (LAB-ID BSIs) in recently discharged patients are likely to be classified as healthcare-associated community-onset (HCA-CO) infections, even though they may represent hospital-onset (HO) infections. A review of LAB-ID BSIs among patients discharged within 14 days revealed that 109 of 756 cases (14.4%) were HO infections. The BSI risk being misclassified as HCA CO may underestimate the hospital infection risk.

Cluster analysis to define distinct clinical phenotypes among septic patients with bloodstream infections.

Prior attempts at identifying outcome determinants associated with bloodstream infection have employed a priori determined classification schemes based on readily identifiable microbiology, infection site, and patient characteristics. We hypothesized that even amongst this heterogeneous population, clinically relevant groupings can be described that transcend old a priori classifications.We applied cluster analysis to variables from three domains: patient characteristics, acuity of illness/clinical presentation and infection characteristics. We validated our clusters based on both content validity and predictive validity.Among 3715 patients with bloodstream infections from Barnes-Jewish Hospital (2008-2015), the most stable cluster arrangement occurred with the formation of 4 clusters. This clustering arrangement resulted in an approximately uniform distribution of the population: Cluster One "Surgical Outside Hospital Transfers" (21.5%), Cluster Two "Functional Immunocompromised Patients" (27.9%), Cluster Three "Women with Skin and Urinary Tract Infection" (28.7%) and Cluster Four "Acutely Sick Pneumonia" (21.8%). Staphylococcus aureus distributed primarily to Clusters Three (40%) and Four (25%), while nonfermenting Gram-negative bacteria grouped mainly in Clusters Two and Four (31% and 30%). More than half of the pneumonia cases occurred in Cluster Four. Clusters One and Two contained 33% and 31% respectively of the individuals receiving inappropriate antibiotic administration. Mortality was greatest for Cluster Four (33.8%, 27.4%, 19.2%, 44.6%; P < .001), while Cluster One patients were most likely to be discharged to a nursing home.Our results support the potential for machine learning methods to identify homogenous groupings in infectious diseases that transcend old a priori classifications. These methods may allow new clinical phenotypes to be identified potentially improving the severity staging and development of new treatments for complex infectious diseases.

Machine learning for fast identification of bacteraemia in SIRS patients treated on standard care wards: a cohort study.

Bacteraemia is a life-threating condition requiring immediate diagnostic and therapeutic actions. Blood culture (BC) analyses often result in a low true positive result rate, indicating its improper usage. A predictive model might assist clinicians in deciding for whom to conduct or to avoid BC analysis in patients having a relevant bacteraemia risk. Predictive models were established by using linear and non-linear machine learning methods. To obtain proper data, a unique data set was collected prior to model estimation in a prospective cohort study, screening 3,370 standard care patients with suspected bacteraemia. Data from 466 patients fulfilling two or more systemic inflammatory response syndrome criteria (bacteraemia rate: 28.8%) were finally used. A 29 parameter panel of clinical data, cytokine expression levels and standard laboratory markers was used for model training. Model tuning was performed in a ten-fold cross validation and tuned models were validated in a test set (80:20 random split). The random forest strategy presented the best result in the test set validation (ROC-AUC: 0.729, 95%CI: 0.679-0.779). However, procalcitonin (PCT), as the best individual variable, yielded a similar ROC-AUC (0.729, 95%CI: 0.679-0.779). Thus, machine learning methods failed to improve the moderate diagnostic accuracy of PCT.

[Acute haematogenous osteomyelitis in children : Diagnostic algorithm and treatment strategies]. / Akute hämatogene Osteomyelitis im Wachstumsalter : Diagnostischer Algorithmus und Behandlungsstrategien.

Acute haematogenous osteomyelitis (AHO) in children is a severe condition. A delay in diagnosis and insufficient treatment may result in deformities, chronicity and sepsis. Therefore a structured diagnostic workup has to be followed in order to diagnose or rule out osteomyelitis. To identify the causative agent for targeted antibiotic treatment, a bone biopsy or puncture should be performed. However, approximately 25% of cases are culture-negative even after biopsy. The knowledge of the typical age-dependent bacterial spectrum is essential for empirical antibiotic therapy. The principal causative organism is Staphylococcus aureus. Surgery is not routinely required in paediatric acute osteomyelitis but surgical intervention is indicated if an abscess is detected. Secondary septic arthritis is a serious complication which has to be treated immediately by surgical intervention. Nevertheless, complete regeneration can be expected in up to 80% of children with AHO.

Características diferenciales en las bacteriemias según la edad en un hospital comunitario / Differential characteristics of bacteraemias according to age in a community hospital

Objetivo. Describir las características de las bacteriemias, según la edad, en un hospital comunitario. Material y método. Estudio prospectivo de las bacteriemias en el año 2011. Los pacientes se clasificaron en 3 grupos de edad: menos de 65, de 65 a 79 y 80 o más años. Se recogieron variables de los pacientes y de los episodios. Resultados. Se analizaron 233 bacteriemias en 227 pacientes (23,8% en<65; 38,3% entre 65 y 79; y 37,9% en≥80 años). La enfermedad de base más frecuente en todos los grupos fue la diabetes mellitus. En los pacientes muy ancianos el índice de Charlson fue mayor, hubo una menor proporción de factores exógenos y casi un 25% eran dependientes graves (índice de Barthel<20). Escherichia coli fue el germen más frecuente y el foco principal fue el urológico. En los pacientes≥80 años predominó el origen de la infección asociado a cuidados sanitarios, la expresividad clínica menos grave (sepsis) (66,3%) y la mortalidad más elevada (29,1%), respecto a los de menor edad. Conclusiones. Los pacientes muy ancianos con bacteriemia presentaron menos factores exógenos, más comorbilidad y una situación funcional peor; el foco más frecuente fue el urológico y el origen el asociado a cuidados sanitarios. A pesar de que su presentación clínica fue menos grave, su mortalidad fue superior, siendo el grado de dependencia una variable de riesgo independiente muy relevante (AU)

Objective. To describe the characteristics of bacteraemias, according to age, in a community hospital. Material and method. A prospective study of bacteraemias was conducted in 2011. The patients were classified into 3 age groups: younger than 65 years, 65 to 79, and 80 or older. The study collected variables on the patients and episodes. Results. The study analysed 233 bacteraemias in 227 patients (23.8% in those younger than 65 years; 38.3% in the 65 to 79 age group; and 37.9% in the 80 years or older group). The most common underlying disease in all the groups was diabetes mellitus. In the most elderly patients, the Charlson index was highest, there was a lower proportion of exogenous factors, and almost 25% were severely dependent (Barthel index<20). Escherichia coli was the most common germ, and the main focus was urological. The patients aged 80 years or older had predominantly healthcare-associated infections, less severe symptoms (sepsis) (66.3%) and higher mortality (29.1%) compared with the younger patients. Conclusions. The very elderly patients with bacteraemia presented fewer exogenous factors, greater comorbidity and a poorer functional situation. The most common focus was urological and the origin was healthcare related. Despite their less severe clinical presentation, these patients’ mortality was greater, and their degree of dependence was a highly relevant independent risk factor (AU)

Blood culture-PCR to optimise typhoid fever diagnosis after controlled human infection identifies frequent asymptomatic cases and evidence of primary bacteraemia.

BACKGROUND: Improved diagnostics for typhoid are needed; a typhoid controlled human infection model may accelerate their development and translation. Here, we evaluated a blood culture-PCR assay for detecting infection after controlled human infection with S. Typhi and compared test performance with optimally performed blood cultures. METHODOLOGY/PRINCIPAL FINDINGS: Culture-PCR amplification of blood samples was performed alongside daily blood culture in 41 participants undergoing typhoid challenge. Study endpoints for typhoid diagnosis (TD) were fever and/or bacteraemia. Overall, 24/41 (59%) participants reached TD, of whom 21/24 (86%) had ≥1 positive blood culture (53/674, 7.9% of all cultures) or 18/24 (75%) had ≥1 positive culture-PCR assay result (57/684, 8.3%). A further five non-bacteraemic participants produced culture-PCR amplicons indicating infection; overall sensitivity/specificity of the assay compared to the study endpoints were 70%/65%. We found no significant difference between blood culture and culture-PCR methods in ability to identify cases (12 mismatching pairs, p = 0.77, binomial test). Clinical and stool culture metadata demonstrated that additional culture-PCR amplification positive individuals likely represented true cases missed by blood culture, suggesting the overall attack rate may be 30/41 (73%) rather than 24/41 (59%). Several participants had positive culture-PCR results soon after ingesting challenge providing new evidence for occurrence of an early primary bacteraemia. CONCLUSIONS/SIGNIFICANCE: Overall the culture-PCR assay performed well, identifying extra typhoid cases compared with routine blood culture alone. Despite limitations to widespread field-use, the benefits of increased diagnostic yield, reduced blood volume and faster turn-around-time, suggest that this assay could enhance laboratory typhoid diagnostics in research applications and high-incidence settings.

The daily risk of bacteremia during hospitalization and associated 30-day mortality evaluated in relation to the traditional classification of bacteremia.

BACKGROUND: We investigated the overall and daily incidence of bacteremia among hospitalized patients and evaluated the traditional classification of bacteremia (community-onset vs nosocomial based on a 48-hour time window) by means of the daily incidence and associated 30-day mortality. METHODS: In a multicenter hospital-based cohort study, we included all patients aged 15 years or older admitted to hospitals in Funen County, Denmark, during 2000-2008, and identified all first bacteremias per admission. We calculated the overall incidence of bacteremia per 1,000 admissions and 10,000 bed-days, as well as the daily incidence of bacteremia per 10,000 bed-days and associated 30-day mortality. RESULTS: We included 724,339 admissions and 10,281 bacteremias for an overall incidence of 14.2 per 1,000 admissions and 23.6 per 10,000 bed-days. The daily incidence was highest on the first 2 days of admission followed by lower incidences that were constant beyond day 12, but varied according to patient and epidemiologic characteristics. Thirty-day mortality for patients with bacteremia was 18% on day 1, 21% on day 2, and between 25% and 35% thereafter. CONCLUSIONS: Hospitalized patients were at the highest risk of bacteremia during the first 2 days followed by lower incidences that were constant beyond day 12. Thirty-day mortality was 18%-21% for patients with bacteremia on the first 2 days and 25%-35% thereafter. Our findings support the traditional classification of bacteremia.

The burden of mucosal barrier injury laboratory-confirmed bloodstream infection among hematology, oncology, and stem cell transplant patients.

OBJECTIVE: To evaluate the impact and burden of the new National Healthcare Safety Network surveillance definition, mucosal barrier injury laboratory-confirmed bloodstream infection (MBI-LCBI), in hematology, oncology, and stem cell transplant populations. DESIGN: Retrospective cohort study. SETTING: Two hematology, oncology, and stem cell transplant units at a large academic medical center. METHODS: Central line-associated bloodstream infections (CLABSIs) identified during a 14-month period were reviewed and classified as MBI-LCBI or non-MBI-LCBI (MBI-LCBI criteria not met). During this period, interventions to improve central line maintenance were implemented. Characteristics of patients with MBI-LCBI and non-MBI-LCBI were compared. Total CLABSI, MBI-LCBI, and non-MBI-LCBI rates were compared between baseline and postintervention phases of the study period. RESULTS: Among 66 total CLABSI cases, 47 (71%) met MBI-LCBI criteria. Patients with MBI-LCBI and non-MBI-LCBI were similar in regard to most clinical and demographic characteristics. Between the baseline and postintervention study periods, the overall CLABSI rate decreased from 3.37 to 3.21 infections per 1,000 line-days (incidence rate ratio, 0.95; 4.7% reduction, P=.84), the MBI-LCBI rate increased from 2.08 to 2.61 infections per 1,000 line-days (incidence rate ratio, 1.25; 25.3% increase, P=.44), and the non-MBI-LCBI rate decreased from 1.29 to 0.60 infections per 1,000 line-days (incidence rate ratio, 0.47; 53.3% reduction, P=.12). CONCLUSIONS: Most CLABSIs identified among hematology, oncology, and stem cell transplant patients met MBI-LCBI criteria, and CLABSI prevention efforts did not reduce these infections. Further review of the MBI-LCBI definition and impact is necessary to direct future definition changes and reporting mandates.

[Classification of haematogenous and post-traumatic osteomyelitis]. / Klassifikation der Osteomyelitis und Osteitis.

A classification of osteomyelitis must reflect the complexity of the disease and, moreover, provide conclusions for the treatment. The classification is based on the following eight parameters: source of infection (OM [osteomyelitis]/OT [post-traumatic OM]), anatomic region, stability of affected bone (continuity of bone), foreign material (internal fixation, prosthesis), range of infection (involved structures), activity of infection (acute, chronic, quiescent), causative microbes (unspecific and specific bacteria, fungi) and comorbidity (immunosuppressive diseases, general and local). In the long version of the classification, which was designed for scientific studies, the parameters are named by capital letters and specified by Arabic numbers, e.g., an acute, haematogenous osteomyelitis of a femur in an adolescent with diabetes mellitus, caused by Staphylococcus aureus, multi-sensible is coded as: OM2 Lo33 S1a M1 In1d Aa1 Ba2a K2a. The letters and numbers can be found in clearly arranged tables or calculated by a freely available grouper on the internet (www.osteomyelitis.exquit.net). An equally composed compact version of the classification for clinical use includes all eight parameters, but without further specification. The above-mentioned example in the compact version is: OM 3 S a Ba2 K2. The short version of the classification uses only the first six parameters and excludes causative microbes and comorbidity. The above mentioned example in the short version is: OM 3 S a. The long version of the classification describes an osteomyelitis in every detail. The complexity of the patient's disease is clearly reproducible and can be used for scientific comparisons. The for clinical use suggested compact and short versions of the classification include all important characteristics of an osteomyelitis, can be composed quickly and distinctly with the help of tables and provide conclusions for the individual treatment. The freely available grouper (www.osteomyelitis.exquit.net) creates all three versions of the classification in one step.

Diagnosis and classification of adult Still's disease.

The cornerstone of adult onset Still's disease is the triad of daily fever, arthritis and rash. This syndrome remains enigmatic and most often a disease of exclusion. There are both musculoskeletal as well as systemic features. More importantly, reactive hemophagocytic syndrome may occur in patients. In this review we attempt to place this syndrome in perspective, including data on geoepidemiology, clinical and laboratory features.

Mucosal barrier injury laboratory-confirmed bloodstream infection: results from a field test of a new National Healthcare Safety Network definition.

OBJECTIVE: To assess challenges to implementation of a new National Healthcare Safety Network (NHSN) surveillance definition, mucosal barrier injury laboratory-confirmed bloodstream infection (MBI-LCBI). DESIGN: Multicenter field test. SETTING: Selected locations of acute care hospitals participating in NHSN central line-associated bloodstream infection (CLABSI) surveillance. METHODS: Hospital staff augmented their CLABSI surveillance for 2 months to incorporate MBI-LCBI: a primary bloodstream infection due to a selected group of organisms in patients with either neutropenia or an allogeneic hematopoietic stem cell transplant with gastrointestinal graft-versus-host disease or diarrhea. Centers for Disease Control and Prevention (CDC) staff reviewed submitted data to verify whether CLABSIs met MBI-LCBI criteria and summarized the descriptive epidemiology of cases reported. RESULTS: Eight cancer, 2 pediatric, and 28 general acute care hospitals including 193 inpatient units (49% oncology/bone marrow transplant [BMT], 21% adult ward, 20% adult critical care, 6% pediatric, 4% step-down) conducted field testing. Among 906 positive blood cultures reviewed, 282 CLABSIs were identified. Of the 103 CLABSIs that also met MBI-LCBI criteria, 100 (97%) were reported from oncology/BMT locations. Agreement between hospital staff and CDC classification of reported CLABSIs as meeting the MBI-LCBI definition was high (90%; κ = 0.82). Most MBI-LCBIs (91%) occurred in patients meeting neutropenia criteria. Some hospitals indicated that their laboratories' methods of reporting cell counts prevented application of neutropenia criteria; revised neutropenia criteria were created using data from field testing. CONCLUSIONS: Hospital staff applied the MBI-LCBI definition accurately. Field testing informed modifications for the January 2013 implementation of MBI-LCBI in the NHSN.

Prediction of bacteremia in children with febrile episodes during chemotherapy for acute lymphoblastic leukemia.

The purpose was to identify risk factors for bacteremia in febrile episodes occurring during chemotherapy for acute lymphoblastic leukemia (ALL) in children, and to develop a risk score permitting risk-adapted antibiotic therapy. We reviewed a total of 172 febrile episodes occurring during chemotherapy in 31 children and adolescents with ALL. Temperature, hematological parameters, culture findings, and antibiotic therapy were recorded. Bacteremias were classified as transmucosal or CVC-dependent. Blood cultures were positive with mucosal pathogens in 15 cases (9%) and with skin/environmental bacteria in 34 (20%). CVC-dependent infections occurred throughout the treatment phases, while transmucosal primarily during induction therapy. Transmucosal bacteremia was associated with induction therapy, leukocyte count ≤0.5 × 10(9)/L, neutrophil count ≤0.1 × 10(9)/L, monocyte count ≤0.01 × 10(9)/L, and platelet count ≤50 × 10(9)/L. Based on logistic conversion of the odds ratios for the five factors, a weight of 2 was assigned to induction therapy and leukocyte count ≤0.5 × 10(9)/L, and a weight of 1 to the remaining three parameters. The weights were included in a simple additive score ranging from 0 to 7, which defined groups with 4%, 6%, 24%, and 40% risk of transmucosal bacteremia. CVC-dependent bacteremia was not associated with markers of poor bone marrow function. In conclusion, transmucosal bacteremia in children with ALL is related to infiltration or suppression of the bone marrow. A score reflecting the condition of the marrow can define low-risk and high-risk groups and may prove clinically useful.

Discord among performance measures for central line-associated bloodstream infection.

BACKGROUND: Central line-associated bloodstream infection (CLABSI) is a national target for mandatory reporting and a Centers for Medicare and Medicaid Services target for value-based purchasing. Differences in chart review versus claims-based metrics used by national agencies and groups raise concerns about the validity of these measures. OBJECTIVE: Evaluate consistency and reasons for discordance among chart review and claims-based CLABSI events. METHODS: We conducted 2 multicenter retrospective cohort studies within 6 academic institutions. A total of 150 consecutive patients were identified with CLABSI on the basis of National Healthcare Safety Network (NHSN) criteria (NHSN cohort), and an additional 150 consecutive patients were identified with CLABSI on the basis of claims codes (claims cohort). All events had full-text medical record reviews and were identified as concordant or discordant with the other metric. RESULTS: In the NHSN cohort, there were 152 CLABSIs among 150 patients, and 73.0% of these cases were discordant with claims data. Common reasons for the lack of associated claims codes included coding omission and lack of physician documentation of bacteremia cause. In the claims cohort, there were 150 CLABSIs among 150 patients, and 65.3% of these cases were discordant with NHSN criteria. Common reasons for the lack of NHSN reporting were identification of non-CLABSI with bacteremia meeting Centers for Disease Control and Prevention (CDC) criteria for an alternative infection source. CONCLUSION: Substantial discordance between NHSN and claims-based CLABSI indicators persists. Compared with standardized CDC chart review criteria, claims data often had both coding omissions and misclassification of non-CLABSI infections as CLABSI. Additionally, claims did not identify any additional CLABSIs for CDC reporting. NHSN criteria are a more consistent interhospital standard for CLABSI reporting.

[Taxonomic composition of metagenomic community in the larval gut of mosquito Anopheles sinensis (Diptera: Culicidae)].

OBJECTIVE: To investigate the bacteria diversity in larval gut of field-collected Anopheles sinensis. METHODS: The 16S rDNA V4 region of An. sinensis larvae collected from paddy on Jiading District of Shanghai (L1/L2) and small seeping water on Wenchang City of Hainan (AS) was sequenced by high-throughput pyrosequencing. Using Qiime and Mothur softwares, the number of sequences and operational taxonomic units (OTUs) for each sample was sorted and calculated, the species abundance and distribution, Alpha diversity index and difference times of species abundance among samples were analyzed. RESULTS: The number of sequences and OTUs for each sample were 253 724/3 930 (L1), 225 203/4 312 (L2) and 73 990/2 380 (AS). The rarefaction curves showed that adequate sampling was achieved. The number of OTUs was close to actual situation. The value of richness index was 5 942.61/6 534.88 (L1), 6 328.17/7 235.89 (L2) and 4228.66/5 651.20 (AS); diversity index was 4.63/0.03 (L1), 5.10/0.02 (L2) and 0.14/3.94 (AS). The dominant species of An. sinensis larvae gut microbiota all belonged to the phylum Proteobacteria, with a percentage of 87% (AS) and 90% (L). In addition, the dominant phyla among them were Firmicutes, Bacteroidetes and Actinobacteria. The comparison of bacterial abundance between L and AS showed that there were 18 phyla with significant difference, except the Proteobacteria and Deinococcus-Thermus; only 9 phyla were different significantly between L1 and L2. CONCLUSION: Evenness and richness of bacteria flora in the An. sinensis larvae gut collected from paddy and small seeping waters were obtained.

Classification of positive blood cultures: computer algorithms versus physicians' assessment--development of tools for surveillance of bloodstream infection prognosis using population-based laboratory databases.

BACKGROUND: Information from blood cultures is utilized for infection control, public health surveillance, and clinical outcome research. This information can be enriched by physicians' assessments of positive blood cultures, which are, however, often available from selected patient groups or pathogens only. The aim of this work was to determine whether patients with positive blood cultures can be classified effectively for outcome research in epidemiological studies by the use of administrative data and computer algorithms, taking physicians' assessments as reference. METHODS: Physicians' assessments of positive blood cultures were routinely recorded at two Danish hospitals from 2006 through 2008. The physicians' assessments classified positive blood cultures as: a) contamination or bloodstream infection; b) bloodstream infection as mono- or polymicrobial; c) bloodstream infection as community- or hospital-onset; d) community-onset bloodstream infection as healthcare-associated or not. We applied the computer algorithms to data from laboratory databases and the Danish National Patient Registry to classify the same groups and compared these with the physicians' assessments as reference episodes. For each classification, we tabulated episodes derived by the physicians' assessment and the computer algorithm and compared 30-day mortality between concordant and discrepant groups with adjustment for age, gender, and comorbidity. RESULTS: Physicians derived 9,482 reference episodes from 21,705 positive blood cultures. The agreement between computer algorithms and physicians' assessments was high for contamination vs. bloodstream infection (8,966/9,482 reference episodes [96.6%], Kappa = 0.83) and mono- vs. polymicrobial bloodstream infection (6,932/7,288 reference episodes [95.2%], Kappa = 0.76), but lower for community- vs. hospital-onset bloodstream infection (6,056/7,288 reference episodes [83.1%], Kappa = 0.57) and healthcare-association (3,032/4,740 reference episodes [64.0%], Kappa = 0.15). The 30-day mortality in the discrepant groups differed from the concordant groups as regards community- vs. hospital-onset, whereas there were no material differences within the other comparison groups. CONCLUSIONS: Using data from health administrative registries, we found high agreement between the computer algorithms and the physicians' assessments as regards contamination vs. bloodstream infection and monomicrobial vs. polymicrobial bloodstream infection, whereas there was only moderate agreement between the computer algorithms and the physicians' assessments concerning the place of onset. These results provide new information on the utility of computer algorithms derived from health administrative registries.