Cox regression with linked data.

Record linkage is increasingly used, especially in medical studies, to combine data from different databases that refer to the same entities. The linked data can bring analysts novel and valuable knowledge that is impossible to obtain from a single database. However, linkage errors are usually unavoidable, regardless of record linkage methods, and ignoring these errors may lead to biased estimates. While different methods have been developed to deal with the linkage errors in the generalized linear model, there is not much interest on Cox regression model, although this is one of the most important statistical models in clinical and epidemiological research. In this work, we propose an adjusted estimating equation for secondary Cox regression analysis, where linked data have been prepared by a third-party operator, and no information on matching variables is available to the analyst. Through a Monte Carlo simulation study, the proposed method is shown to lead to substantial bias reductions in the estimation of the parameters of the Cox model caused by false links. An asymptotically unbiased variance estimator for the adjusted estimators of Cox regression coefficients is also proposed. Finally, the proposed method is applied to a linked database from the Brest stroke registry in France.

Apparent Absence of Selective Pressure on Pneumocystis jirovecii Organisms in Patients with Prior Methotrexate Exposure.

Pneumocystis jirovecii infections occur in patients treated with methotrexate (MTX) because of immunosuppressive effects of this highly potent dihydrofolate reductase (DHFR) inhibitor. Conversely, MTX may act as an anti-P. jirovecii drug and consequently may exert a selective pressure on this fungus. In this context, we compared the sequences of the dhfr gene of P. jirovecii isolates obtained from two groups of patients with P. jirovecii infections. The first group, with systemic diseases or malignancies, had prior exposure to MTX (21 patients), whereas the second group (22 patients), the control group, did not. Three single nucleotide polymorphisms (SNPs) were observed at positions 278, 312, and 381. The first one was located in the intronic region and the two others were synonymous. Based on these SNPs, three P. jirovecii dhfr alleles, named A, B, and C, were specified. Allele A was the most frequent, as it was observed in 18 patients (85.7%) and in 16 patients (72.7%) of the first and second groups, respectively. No significant difference in P. jirovecii dhfr gene diversity in the two patient groups was observed. In conclusion, these original results suggest that MTX does not exert an overt selective pressure on P. jirovecii organisms.

Effectiveness of angiotensin-converting enzyme inhibitors and angiotensin receptor blockers on total and cardiovascular mortality and morbidity in primary prevention: A nationwide study based on French Health Insurance Data (SNDS).

Angiotensin-converting enzyme (ACE) inhibitors and angiotensin II receptor blockers (ARBs) both inhibit the renin-angiotensin system (RAS) but have different sites of action. Whether clinically meaningful differences exist is still debated. The authors set up a population-based nationwide retrospective cohort study with at least 5 years of follow-up based on the comprehensive French Health Insurance Database linked to the French hospital discharge database. Patients aged 50 or above, identified as ARB or ACE inhibitor new users in 2009 (at least one delivery during the year and no such delivery in 2008) were eligible. Exclusion criteria included history of cancer, cardiovascular disease, or chronic renal insufficiency. Main outcome measure was overall mortality. Secondary outcomes were cardiovascular deaths, major cardiovascular events, and major or other cardiovascular events. Out of 407 815 eligible patients, 233 682 (57%) were ARB users; two-third had no previous exposure to antihypertensive drug. Based on propensity-score based Cox model, ARB new user group had a better overall (HR: .878, 95%CI, .854 to .902), and cardiovascular (HR: .841, 95%CI, .800 to .84) survival and had a lower risk for major cardiovascular events (HR: .886, 95%CI, .868 to .905). Statistically significant quantitative interactions were detected with diabetes. Considering subgroup analyses, ARBs had a better survival than ACE inhibitors in nondiabetic patients.

Hospitalization for adverse events under abiraterone or enzalutamide exposure in real-world setting: A French population-based study on prostate cancer patients.

AIMS: Safety profiles of abiraterone and enzalutamide rely mainly on Phase III clinical trials. Our objective was to estimate the incidence rate ratio (IRR) for certain adverse events leading in real life to hospitalization (atrial fibrillation, acute heart failure, ischaemic heart disease, acute kidney injury [AKI], ischaemic stroke, torsade de pointe/QT interval prolongation, hepatitis and seizure), comparing abiraterone to enzalutamide. We also set out to discuss previously identified safety signals. METHOD: Using the French National Health Insurance System database, all patients newly exposed to abiraterone or enzalutamide between 2013 and 2017 and followed until 31 December 2018 were targeted. IRRs for each event were estimated using a Poisson model in a sub-population of patients without contraindications or precautions for use for either treatment. RESULTS: Among 11 534 new users of abiraterone and enzalutamide, AKI (IRR 1.42, 95% CI: 1.01-2.00), liver monitoring suggestive of hepatic damage (IRR 3.06, 95% CI: 2.66-3.53) and atrial fibrillation (IRR 1.12, 95% CI: 1.05-1.19) were significantly more often observed with abiraterone than with enzalutamide. CONCLUSION: Our study provides knowledge on abiraterone and enzalutamide real-life safety profiles, especially for events leading to hospitalization. Despite several limitations, including the lack of clinical data, the safety signal for AKI under abiraterone is in line with results of an analysis of the French pharmacovigilance database, which requires further specific investigations. Enlightening the clinicians' therapeutic choices for patients treated for prostate cancer, our study should lead to clinicians being cautious in the use of abiraterone.

Overall Survival Among Chemotherapy-Naive Patients With Castration-Resistant Prostate Cancer Under Abiraterone Versus Enzalutamide: A Direct Comparison Based on a 2014-2018 French Population Study (the SPEAR Cohort).

Abiraterone acetate (ABI) and enzalutamide (ENZ) are considered to be clinically relevant comparators among chemotherapy-naive patients with castration-resistant prostate cancer. No clinical trials comparing overall survival with ABI versus ENZ in a head-to-head approach have been published so far. A few observational studies with low power suggested a potential benefit of ENZ. We used the French National Health Data System to compare overall survival of new users of ABI and ENZ among chemotherapy-naive patients with castration-resistant prostate cancer in 2014-2017, followed through 2018 (the SPEAR cohort, a 2014-2018 cohort study). With an intent-to-treat approach, a survival analysis was performed, estimating hazard ratios for overall survival with the inverse probability weighted Cox model method. Among 10,308 new users, 64% were treated with ABI and 36% with ENZ. The crude mortality rate was 25.2 per 100 person-years (95% confidence interval (CI): 24.4, 26.0) for ABI and 23.7 per 100 person-years (95% CI: 22.6, 24.9) for ENZ. In the weighted analysis, ENZ was associated with better overall survival compared with ABI (hazard ratio = 0.90 (95% CI: 0.85, 0.96) with a median overall survival of 31.7 months for ABI and 34.2 months for ENZ). When restricting to 2015-2017 new users, the effect estimate shifted up to a hazard ratio of 0.93 (95% CI: 0.86, 1.01).

Patients with Metastatic Melanoma Receiving Anticancer Drugs: Changes in Overall Survival, 2010-2017.

Immune checkpoint inhibitors and targeted therapies have profoundly altered the management of several cancers over the past decade. Metastatic melanoma has been at the forefront of these changes. We provide here a nationwide overview and an assessment of changes in survival in France. We included 10,936 patients receiving a systemic treatment for metastatic cutaneous melanoma between 2010 and 2017 using the French National Health Insurance database (Système National des Données de Santé). Over the study period, there was a doubling of the number of new patients receiving a systemic treatment. Cytotoxic chemotherapy was progressively replaced by targeted therapy and immune checkpoint inhibitors. Patients having initiated a first-line treatment since June 2015 gained 46% overall survival compared with those initiating treatment before 2012. Overall survival at 24 months rose from 21% to 44%. We provide real-world evidence for the improvement of overall survival in the past decade among patients with metastatic melanoma. Although the characteristics of the patients treated can vary across periods, this type of exhaustive real-world data provides evidence from broader populations than those included in clinical trials.

Association Between Early Severe Cardiovascular Events and the Initiation of Treatment With the Anti-Interleukin 12/23p40 Antibody Ustekinumab.

Importance: Ustekinumab, a monoclonal antibody targeting interleukin 12/23p40 (IL-12/23p40), is effective in the treatment of moderate to severe psoriasis, psoriatic arthritis, and Crohn disease. In 2011, a meta-analysis of randomized clinical trials reported a potential risk of severe cardiovascular events (SCEs) within the first few months after the initiation of anti-IL-12/23p40 antibodies. Objective: To assess whether the initiation of ustekinumab treatment is associated with increased risk of SCEs. Design, Setting, and Participants: This case-time-control study used data from the French national health insurance database, covering 66 million individuals, on all patients exposed to ustekinumab between April 1, 2010, and December 31, 2016, classified according to their cardiovascular risk level (high- and low-risk strata). The risk period was the 6 months before the SCE, defined as acute coronary syndrome or stroke, and the reference period was the 6 months before the risk period. Statistical analysis was performed from September 20, 2017, to July 6, 2018. Exposure: The initiation of ustekinumab treatment was screened during the risk and reference periods. Main Outcomes and Measures: Odds ratios for the risk of SCE after the initiation of ustekinumab treatment were calculated. Results: Of the 9290 patients exposed to ustekinumab (4847 men [52%]; mean [SD] age, 43 [14] years), 179 experienced SCEs (65 cases of acute coronary syndrome, 68 cases of unstable angina, and 46 cases of stroke). Among patients with a high cardiovascular risk, a statisically significant association between initiaton of ustekinumab treatment and SCE occurrence was identified (odds ratio, 4.17; 95% CI, 1.19-14.59). Conversely, no statistically significant association was found among patients with a low cardiovascular risk (odds ratio, 0.30; 95% CI, 0.03-3.13). Conclusions and Relevance: This study suggests that the initiation of ustekinumab treatment may trigger SCEs among patients at high cardiovascular risk. In line with the current mechanistic models for atherosclerotic disease, the period after the initiation of anti-IL-12/23p40 may be associated with atherosclerotic plaque destabilization via the inhibition of helper T cell subtype 17. Although the study interpretation is limited by its observational design, these results suggest that caution may be needed in the prescription of ustekinumab to patients at high cardiovascular risk.

Risk of suicide attempt associated with isotretinoin: a nationwide cohort and nested case-time-control study.

BACKGROUND: Isotretinoin is the only effective treatment for severe acne. An isotretinoin-related suicide risk is still debated and under scrutiny by regulatory agencies. Our objectives were: to assess the risk of suicide attempt before, during and after isotretinoin treatment; to detect any potential triggering effect of isotretinoin initiation on suicide attempt. METHODS: We implemented a cohort and nested case-time-control study of subjects treated with oral isotretinoin (course or initiation) aged 10-50 years, using the Nationwide French Health Insurance data (2009-2016). The main outcome was hospitalized suicide attempt. Standardized incidence ratios for hospitalized suicide attempts were calculated before, during and after isotretinoin treatment. The number of isotretinoin initiations was compared in risk and control periods of 2 months using a case-time-control analysis. RESULTS: In all, 443 814 patients (median age 20.0 years; interquartile range 17.0-27.0 years) were exposed to isotretinoin, amounting to 244 154 person-years, with a marked seasonality for treatment initiation. Compared with the French general population, the occurrence of suicide attempts under isotretinoin treatment was markedly lower, with a standardized incidence ratio of 0.6 [95% confidence interval (CI) = 0.53-0.67]; the same applied, to a lesser extent, before and after isotretinoin treatment. In the case-time-control analysis, among cases of suicide attempt, 108 and 127 isotretinoin initiations were observed in the risk and control periods respectively (i.e. 0-2 months and 2-4 months before the date of suicide attempt). The comparison with the 1199 and 1253 initiations observed among matched controls in the same two periods yielded a case-time-control odds ratio of 0.89 (95% CI = 0.68-1.16). A sensitivity analysis using three-month periods and a complementary analysis adding completed suicides for case definition showed consistent results. CONCLUSION: Compared with the general population, a lower risk of suicide attempt was observed among patients exposed to isotretinoin and there was no evidence for a triggering effect of isotretinoin initiation on suicide attempt. A selection of patients at lower risk for suicidal behaviour and appropriate treatment management could explain these findings. Risk management plans should therefore be maintained.

French administrative health care database (SNDS): The value of its enrichment.

SNIIRAM/SNDS, the French administrative health care database, covers around 99% of the population. Its main limitation is the absence of clinical information and biological results. This report exposes the value of SNIIRAM/SNDS enrichment by external databases, and the linkage issues. It is illustrated by examples: the well-known population-based cohort CONSTANCES created to answer to epidemiological research questions with a specific interest on occupational and social factors, chronic diseases, and aging; the CANARI study, a regional-based study that collected Gleason score in all pathology laboratories in Brittany and then, linked pathology results to an ad hoc extraction from SNIIRAM database; the goal was to investigate the risk of high grade prostate cancer in patients treated by 5-alpha-reductase inhibitors for a symptomatic benign prostatic hyperplasia; the SACHA study, that identified and medically validated major bleeding event referred to emergency wards, then linked those clinical data to SNIIRAM; the goal was to minimize misclassification bias when estimating bleeding risk in patients who were prescribed antithrombotic drugs; the ISO-PSY study linked the SNIIRAM with the national cause of death registry (CépiDc) and the nationwide emergency department surveillance system (OSCOUR® network) to investigate the potential link between isotretinoin and suicidal risk; the EFEMERIS cohort that assesses drugs prescriptions in French pregnant women who delivered in the Haute-Garonne region; the EPI-GETB-AM study that derived a SNIIRAM/SNDS-based algorithm to identify venous thromboembolism and linked SNIIRAM/SNDS to the EPI-GETBO-III survey for validation. Another perspective of SNDS enrichment is clinical trials' data for medico-economic assessment, and extended follow-up without attrition bias. Linkage is not straightforward. Apart from regulatory approbation and authorized data center issues, which could be solved by the Health Data Hub Initiative, a multidisciplinary team with medical, pharmacological and methodological knowledge, as well as with technical skills is essential to handle the whole process.

Safety of Oral Propranolol for Infantile Hemangioma.

OBJECTIVES: The safety of oral propranolol for infantile hemangioma has not yet been studied at population level since the pediatric use marketing authorization was obtained in Europe. METHODS: A survey of a nationwide, claim-based observational cohort of children <3 years old, with at least 1 delivery of oral propranolol between July 2014 and June 2016, was performed by using the database of the French National Health Insurance system. Standardized morbidity ratios (SMRs) were calculated by using, from the same database, a representative random sample of nonexposed subjects. The main outcomes were hospitalizations for cardiovascular (conduction disorders, bradycardia, and hypotension), respiratory (bronchial hyperactivity and bronchospasm), or metabolic events (hypoglycemia and hyperkalaemia), identified through the hospitalization diagnostic codes of the International Classification of Diseases, 10th Revision. The main analysis was conducted separately on "healthy" children (N = 1484), that is, free from of any prespecified underlying disease and on children with 1 of these underlying diseases (N = 269). RESULTS: In all, 1753 patients <3 years of age had at least 2 deliveries of oral propranolol. In the healthy population, we observed 2 cardiovascular events (SMR = 2.8 [0-6.7]), 51 respiratory events (SMR = 1.7 [1.2-2.1]), and 3 metabolic events (SMR = 5.1 [0-10.9]). In the population with an underlying disease (mainly congenital heart disease), we observed 11 cardiovascular events leading to an SMR of 6.0 (2.5-9.6). SMRs were not significantly raised for respiratory or metabolic events in this "nonhealthy" population. CONCLUSIONS: In this study on a large continuous nationwide claims database, we confirm the safety profile of oral propranolol in healthy children to be good.

A visual approach of care pathways from the French nationwide SNDS database - from population to individual records: the ePEPS toolbox.

Secondary use of medical and administrative databases has become a powerful tool for epidemiological studies. In that respect, the recent access opening of French nationwide health record database or Système National des Données de Santé is a great opportunity to carry out comprehensive health studies at the country level. However, using this database is far from being straightforward for nonexpert data scientists; so, dedicated tools needed to be developed. Our contribution is focused on data management and visualization tools able to query and cope with the complexity of care pathways of patients. The scope of the toolbox presented below goes from interactive tools for building groups of patients to timeline representations of individual patient healthcare trajectories.

From medico-administrative databases analysis to care trajectories analytics: an example with the French SNDS.

Medico-administrative data like SNDS (Système National de Données de Santé) are not collected initially for epidemiological purposes. Moreover, the data model and the tools proposed to SNDS users make their in-depth exploitation difficult. We propose a data model, called the ePEPS model, based on healthcare trajectories to provide a medical view of raw data. A data abstraction process enables the clinician to have an intuitive medical view of raw data and to design a study-specific view. This view is based on a generic model of care trajectory, that is a sequence of time stamped medical events for a given patient. This model is combined with tools to manipulate care trajectories efficiently.

Androgen deprivation therapy and cardiovascular risk: No meaningful difference between GnRH antagonist and agonists-a nationwide population-based cohort study based on 2010-2013 French Health Insurance data.

BACKGROUND: Observational studies suggested that androgen deprivation therapy (ADT) is associated with an increased cardiovascular (CV) risk. They all compared ADT-treated cancer patients to non-treated patients or non-cancer subjects. Our aim was to evaluate whether CV risk differs by type of ADT. METHODS: Through nationwide population-based claims reimbursement database linked to hospital discharge database, we identified adult men with prostate cancer who initiated ADT (gonadotrophin releasing hormone [GnRH] agonist or antagonist, antiandrogen [AA], combined androgen blockade [CAB]) or had orchiectomy (OT) between 1st July, 2010, and the 31st December, 2011, and followed them up to 31st December, 2013. The main analysis followed an 'on-treatment' approach that censored all patients at the time of first therapeutic modification; it used Cox regression analysis to estimate hazard ratios (HRs) for hospitalisations for ischaemic events (myocardial infarction or ischaemic stroke, whichever came first), adjusted on age, baseline co-morbidities and taking into account death as a competing risk. RESULTS: Among the 35,118 new ADT users, 71% received GnRH agonist (reference group), 12% CAB, 13% AA, 3.6% GnRH antagonist and 0.6% had OT. CAB was associated with an increased risk (adjusted HR [95% confidence interval {CI}], 1.6 [1.3-2.0]) and AA with a decreased risk (adjusted HR [95% CI], 0.6 [0.4-0.9]) of ischaemic events when compared to GnRH agonist. No significant association was found with GnRH antagonist (adjusted HR [95% CI], 1.2 (0.7-2.1)). CONCLUSION: CV risk appeared different across ADT modalities. The probability of a clinically meaningful difference when comparing GnRH antagonists to agonists appears rather low. In a context where better overall and cancer specific survival without worsening quality of life is a challenge for clinicians, a potential heterogeneity in CV morbidity becomes crucial when choosing an ADT.

Safety of Fixed Dose of Antihypertensive Drug Combinations Compared to (Single Pill) Free-Combinations: A Nested Matched Case-Control Analysis.

To compare serious adverse events of fixed-dose dual antihypertensive drug combination (FIXED) to component-based free-combination (FREE).A population-based nationwide cohort from the French Health Insurance System included subjects over 50 years with first time claims (new user) in the second half of 2009 for a calcium-channel blocker or a thiazide-like diuretic in combination with either an angiotensin-converting enzyme inhibitor or an angiotensin receptor blocker as FREE or FIXED. We designed a nested matched case-control analysis with 304 cases, hospitalized for hypotension, syncope, or collapse (n = 224), renal failure (n = 19), hyponatremia, hyper- or hypokalemia (n = 61) and 1394 controls matched for gender, age, date of inclusion in the cohort, and administrative county. Subjects with a medical history of cardiovascular disease, chronic renal failure, or cancer were excluded.The mean age ±â€ŠSD was 73 ±â€Š10 years and 70% were women. Based on the last delivery preceding the index date, 1414 patients (83%) were exposed to FIXED. Homogeneity of FIXED effect compared to FREE across components of the main composite outcome was rejected (P = 0.0099). FIXED formulation significantly increased the odd of the most frequent component (ie, hypotension, syncope, or collapse): OR = 1.88 (95% CI: 1.15-3.05) compared to FREE after adjusting for confounding factors including dose.Serious adverse event occurring in the early phase of treatment deserves attention of physicians because it could alter the benefit/risk ratio of antihypertensive drug combination.

Brand name to generic substitution of antiepileptic drugs does not lead to seizure-related hospitalization: a population-based case-crossover study.

PURPOSE: There is still controversy on brand-to-generic (B-G) antiepileptic drugs (AEDs) substitution. METHODS: To assess association between B-G AED substitution and seizure-related hospitalization, we designed a case crossover using the French National Health Insurance Database. We identified a cohort of adult patients who filled a prescription in 2009-2011 for AEDs with at least one brand name and one generic form. The outcome date was defined as the date of hospitalization, coded G40.x or G41.x, with a G40/G41 hospitalization-free period of at least 1 year. Patients with a medical history of cancer and women who gave birth in 2009-2011 were excluded. We required individuals to have regular dispensations of AEDs within the year preceding the outcome date. Free patients were defined as patients who had only brand name dispensations before the control period. RESULTS: Eight thousand three hundred seventy nine patients (mean age ± standard deviation, 52.7 ± 18.8 years; sex ratio male/female, 1.27) were analyzed. Discordant pairs were 491 with B-G substitution in the control period only and 478 with B-G substitution in the case period only; odds ratio (95% confidence interval) 0.97 (0.86-1.10). No statistically significant interaction was detected among the four prespecified subgroup analyses (gender, age strata, free or non-free, and strict AED monotherapy or not). Controlling for non-seizure-related hospitalizations made no material difference. Sensitivity analyses yielded similar results. CONCLUSIONS: Brand-to-generic AED substitution was not associated with an elevated risk of seizure-related hospitalization.

Roogle: an information retrieval engine for clinical data warehouse.

High amount of relevant information is contained in reports stored in the electronic patient records and associated metadata. R-oogle is a project aiming at developing information retrieval engines adapted to these reports and designed for clinicians. The system consists in a data warehouse (full-text reports and structured data) imported from two different hospital information systems. Information retrieval is performed using metadata-based semantic and full-text search methods (as Google). Applications may be biomarkers identification in a translational approach, search of specific cases, and constitution of cohorts, professional practice evaluation, and quality control assessment.

Design of an automatic coding algorithm for a multi-axial classification in pathology.

An algorithm for automatic coding of pathology reports using a multi-axial codification (ADICAP) is described and evaluated. It extracts << significant words >> or expressions from a corpus and records the statistical relationships between them and the modalities of the different axes. Different weighting functions are evaluated. With the best settings, in more than two cases out of three the correct modality was found among the top 5 list of candidates, except for the << organ >> axis. Several ways of improvement are discussed especially regarding the poor results on the << organ >> axis. Perspectives of a two stages assembling algorithm completing this first step are proposed.

Automatic concept extraction from spoken medical reports.

OBJECTIVE: The objective of this project is to investigate methods whereby a combination of speech recognition and automated indexing methods substitute for current transcription and indexing practices. METHODS: We based our study on existing speech recognition software programs and on NOMINDEX, a tool that extracts MeSH concepts from medical text in natural language and that is mainly based on a French medical lexicon and on the UMLS. For each document, the process consists of three steps: (1) dictation and digital audio recording, (2) speech recognition, (3) automatic indexing. The evaluation consisted of a comparison between the set of concepts extracted by NOMINDEX after the speech recognition phase and the set of keywords manually extracted from the initial document. The method was evaluated on a set of 28 patient discharge summaries extracted from the MENELAS corpus in French, corresponding to in-patients admitted for coronarography. RESULTS: The overall precision was 73% and the overall recall was 90%. Indexing errors were mainly due to word sense ambiguity and abbreviations. A specific issue was the fact that the standard French translation of MeSH terms lacks diacritics. A preliminary evaluation of speech recognition tools showed that the rate of accurate recognition was higher than 98%. Only 3% of the indexing errors were generated by inadequate speech recognition. DISCUSSION: We discuss several areas to focus on to improve this prototype. However, the very low rate of indexing errors due to speech recognition errors highlights the potential benefits of combining speech recognition techniques and automatic indexing.

Combining voice recognition and automatic indexing of medical reports.

Medical records have been evolving from the traditional paper-based records to digital ones, from the method of dictating reports and transcription to voice recognition systems. The transition to digital operations will not be complete until we have the ability to combine voice recognition with automated indexing of texts. This paper introduces the methods we used to evaluate existing voice recognition software programs and presents NOMINDEX, a system that turns a medical text into MeSH codes, using the French ADM lexical database. Those systems were applied to 28 patient discharge summaries in French, produced after a coronarography, and extracted from the MENELAS corpus of texts. Using the best configuration for voice recognition, the rate of accurate recognition exceeds 98 percent. Among the indexing concepts assigned by NOMINDEX, 25 percent were not pertinent and 12 percent of the relevant concepts were missing. Most errors were related to confusion between common language and medical language, and to the coverage of the ADM lexical database. Best results would be expected with a more comprehensive lexical resource In addition, only 3 percent of the errors generated by inadequate voice recognition that remained in the configuration that performed better, impacted on automatic indexing by NOMINDEX.