A preprocessing method for improving data mining techniques. Application to a large medical diabetes database.

The Knowledge Discovery in Databases (KDD) methodology seems to be attractive on the analyze of large clinical databases. In the KDD process, the preprocessing step (data cleaning and handling of missing values) is paramount since it conditions the quality of the results obtained by data mining procedures and represents about 80% of the whole project time. The aims of the present study were to analyze this step and provide tools to handle inconsistent data and missing values. We have broken down the process into 3 main stages: data cleaning--explanatory study of missing values--choice of the procedure used for handling missing values. The data cleaning stage was based on a system of logical rules to correct mistakes and on cluster analysis to discard the poorly filled files. The missing-data mechanism was analyzed by means of multivariate statistical procedures. Two methods to deal with missing values were compared: imputation by the most common value (mode) and imputation using decision trees. This study was performed on a large medical diabetes database (23,601 patients) including numerous missing values. A system of logical rules allowed to correct mistakes on essential parameters (for example, the type of diabetes). Cluster analysis allowed to identify 10% of poorly filled files. After multivariate analysis, the missing-data mechanism could be considered as random. For variables with low number of missing values (< 10%) and categories (< 4), imputation using decision trees provided better results than imputation by mode.

Pregnancy induced hypertension in women with gestational carbohydrate intolerance: the diagest study.

OBJECTIVE: To determine the relationship between pregnancy induced hypertension (PIH) and gestational glucose intolerance. METHODS: A 50g, 1h glucose loading test was offered to all pregnant women between 24 and 28 weeks of gestation in 15 centres in northern France during 8 months in 1992. If the test was positive (> or =7.2 mmol/l), the woman underwent a 3h oral glucose tolerance test (OGTT) as soon as possible. Using the criteria of Carpenter and Coustan, gestational diabetes mellitus (GDM) was defined by two abnormal values (n=218) and gestational mild hyperglycemia (GMH) by one abnormal value (n=130). Each control group was defined by a 50g, 1h loading test result of <7.2 mmol/l (n=108).PIH included gestational hypertension (GH) and preeclampsia (PE). GH was defined as a diastolic pressure of more than 85 mmHg on at least two occasions arising during pregnancy. PE was defined as GH with proteinuria > or =500 mg/24h. RESULTS: The rate of PIH in the three groups (GDM; GMH and control group, C) was, respectively 17.0, 10.8, and 4.6%. All the six PE occurred in the GDM group. Univariate analysis showed significantly higher rate of hypertension in women with a history of PE, increasing body mass index before pregnancy (BMI) and glucose intolerance. In multivariate analysis with adjustment for primiparity, independent risk factors for PIH were a history of PE, BMI>27 and GDM, contrary to GMH and maternal age. CONCLUSIONS: PIH appears to be linked to the level of glucose intolerance during pregnancy, independently of other known factors of hypertension.

Higher carbohydrate intake is associated with decreased incidence of newborn macrosomia in women with gestational diabetes.

OBJECTIVE: To study the influence of energy and macronutrient intake on infant birthweight in women with gestational diabetes mellitus undergoing intensive management. DESIGN: This prospective study evaluated the impact of intensive management of gestational diabetes on maternal and fetal morbidity, and addressed the relationship between food intake and infant birthweight. SETTING: Fifteen maternity hospitals in northern France. SUBJECTS: Ninety-nine women with gestational diabetes or gestational mild hyperglycemia diagnosed between 24 and 34 weeks of gestation were surveyed. After 1 was excluded because of a premature birth and 18 were excluded as underreporters, 80 women were included in the final analysis. Diet intake was assessed by a dietary history at the first interview, and by two 3-day diet records at the 3rd and 7th week after diagnosis. RESULTS: In a forward-stepwise regression analysis (controlling for maternal age; smoking; parity; prepregnancy BMI; pregnancy weight gain; gestational duration; infant sex; fasting and 2-hour postprandial serum glucose; insulin therapy; and energy, fat, protein and carbohydrate intake during treatment) infant birthweight was positively associated with gestational duration (beta = +0.34, P<.002), and negatively with smoking (beta = -0.27, P<.02) and carbohydrate intake (beta = -0.24, P<.03). There were no large-for-gestational-age infants among women whose carbohydrate intake exceeded 210 g/day. CONCLUSION: For women with gestational diabetes undergoing intensive management, higher carbohydrate intake is associated with decreased incidence of macrosomia. APPLICATION: These findings suggest that nutrition counseling in gestational diabetes must be directed to maintain a sufficient carbohydrate intake (at least 250 g per day), which implies a low-fat diet to limit energy intake. A careful distribution of carbohydrate throughout the day and the use of low-glycemic index foods may help limit postprandial hyperglycemia.

Social and health status of arrivals in a French prison: a consecutive case study from 1989 to 1995.

BACKGROUND: To assess the demographic, socioeconomic and health status of male arrivals in French jails and to analyze the time trends of these characteristics. METHODS: The study was carried out in a prison for detained persons and short term prisoners. Using a standardized questionnaire, we recorded the characteristics of all male detainees and prisoners arriving in the prison between 1989 and 1995. The information collected concerned: demographic data, level of education and professional status, reasons for detention or imprisonment, social and family background, lifestyle, medical and psychiatric history, suicide attempts and illicit use of drugs. The characteristics of the study population were compared with those found in the general regional population. We analyzed developing trends in the health status of the prison population as well as their socio-economic profile over a period of seven years (1989 to 1995). RESULTS: A total of 14,785 questionnaires were analyzed. Of the study population, 56% had no professional qualification, and 62% was unemployed. About two-thirds of the inmates presented psychiatric problems or problems with illicit drug use (one or several drugs). Amongst these, 70% had not had any form of care -neither therapeutic nor educational- before their arrival in prison. Between 1989 and 1995, the proportion of drug users increased by 30%. A parallel increase was observed in the level of unemployment and in the frequency of mental problems. CONCLUSIONS: Our results suggest a need for therapeutic and educational care to be provided for prison inmates. This poses a problem which needs to be addressed in terms of public health. The study also illustrates the usefulness of a standardised questionnaire for each arrival. The systematic use of such a tool would make it possible to identify inmates'needs and to propose adapted care solutions.

Is mild gestational hyperglycaemia associated with maternal and neonatal complications? The Diagest Study.

AIMS: To evaluate the maternal and neonatal complications rates of mild gestational hyperglycaemia (MGH) compared to a control group in France. METHODS: A systematic screening by a 50-g glucose challenge test was offered to all women between 24 and 28 weeks of gestation in 15 maternity units. If the 50-g glucose challenge test was > or = 7.2 mmol/l, a 100-g 3-h oral glucose tolerance test (OGTT) was performed. MGH (n = 131) was defined by one abnormal value on the 3-h OGTT (Carpenter and Coustan criteria). The control group (n = 108) was defined by a 50-g glucose challenge test below 7.2 mmol/l. Women with MGH received no treatment or specific advice during the pregnancy. Large for gestational age (LGA) was defined by a birth weight of at least the 90th percentile on French standard growth curves. RESULTS: Women with MGH were older than the controls (28.8 (5.8) vs. 27.0 (5.2); P < 0.05) and had a higher body mass index (24.8 (4.8) vs. 23.0 (3.9); P < 0.01). The rate of pregnancy-induced hypertension and Caesarean section were not different between the MGH and control group. The rate of LGA was significantly higher in the MGH group than the control group (22.1% vs. 11.4%; P < 0.05). After adjustment for confounding factors of macrosomia (pre-pregnancy body mass index > 27, maternal age > 35, multiparity and educational level), there was a persistent relationship between LGA and MGH (odds ratio 2.50; 95% confidence interval (1.16-5.40); P < 0.05). MGH was more frequently associated with adverse maternal and fetal outcome than in the controls (53.4% vs. 28.7%; P < 0.01). CONCLUSIONS: This study suggested that the increased rate of adverse maternal and fetal outcome, especially LGA, was associated with untreated mild gestational hyperglycaemia women compared to a control group. This link to lower degrees of hyperglycaemia during pregnancy is independent of confounding factors.

A prototype of an information system for assessing the health status of prison inmates.

A research-action program was established in 1996 between the Loos-Lez-Lille prison psychiatric unit and the Department of Medical Informatics of the University Hospital of Lille (France):--(1) to investigate the health status and the general characteristics of the prison population--(2) to develop an Information System for improving the prison health care and to facilitate social rehabilitation of convicts. Starting off 1988, all new prisoners are interviewed on their arrival using a standard questionnaire. The transfer of all the information recorded in this questionnaire into a computer base was initiated in 1996, when the research action program began. A statistical analysis was performed on 15,200 records (1989-1995) to identify the most informative parameters: 50% of inmates were less than 24 years old; 57% were unemployed; 60% had no professional qualification. 31% of inmates had a psychiatric history and 16% had made a previous suicide attempt. The rate of drug abuse has increased from 24% in 89 to 53% in 95. To analyze the time trends of these parameters, a prototype of Information System was then developed. The system uses the database to product standard reports in real time.

Comparison between fat intake assessed by a 3-day food record and phospholipid fatty acid composition of red blood cells: results from the Monitoring of Cardiovascular Disease-Lille Study.

We investigated the relationship between assessment of fatty acid intake by a 3-day food record and by capillary gas chromatography of erythrocyte phospholipid fatty acid. The study was performed in a sample of 244 men aged 45 to 66 years from the general population who were participating in the Monitoring of Cardiovascular Disease (MONICA)-Lille survey. The relationship between each nutrient and food item and erythrocyte phospholipid fatty acid was investigated by a regression model on proportion including each food item and nutrient as a dependent variable and percentage of fatty acid and covariables (nonalcoholic energy intake, age, alcohol intake, and smoking) as independent variables. Polyunsaturated fat and linoleic acid intake were positively correlated with linoleic acid content of erythrocytes (beta = 0.641 and 0.604, respectively, P < .001). Monounsaturated and saturated fat intake were correlated with oleic acid (beta = 0.375 and 0.373, respectively, P < .01). Fish intake correlated positively with docosahexaenoic acid (DHA) (beta = 0.383, P < .001) and negatively with arachidonic acid (beta = -0.509, P < .01). These data confirm, on a group level, a good relationship between assessment of polyunsaturated fat intake by a 3-day record and linoleic acid content of erythrocyte membranes. These data suggest that erythrocyte oleic acid content is a marker of both saturated and monounsaturated fat intake.

Incidence of inflammatory bowel disease in northern France (1988-1990).

There were no data concerning the incidence of inflammatory bowel disease (IBD) in France. The aim of this study was to investigate the incidence of Crohn's disease and ulcerative colitis in northern France. This prospective population based study was realised through the gastroenterologists of the region Nord-Pas de Calais and the Somme Department. Each gastroenterologist referred patients consulting for the first time with clinical symptoms compatible with IBD. Data were collected by an interviewer practitioner present at the gastroenterologist's consulting room. Two independent expert gastroenterologists assessed each case in a blind manner and made a final diagnosis of Crohn's disease, ulcerative colitis, ulcerative proctitis, or unclassifiable chronic colitis. From 1988 to 1990, 1291 cases of IBD were recorded: 674 (52%) Crohn's disease, 466 (36%) ulcerative colitis including 162 proctitis (35%), and 151 (12%) unclassifiable chronic colitis. The mean annual incidence was 4.9 per 100,000 for Crohn's disease and 3.2 for ulcerative colitis. The sex ratio F/M was 1.3 for Crohn's disease and 0.8 for ulcerative colitis. The highest age specific incidence rate for Crohn's disease was between 20 and 29 years: 13.1 for women and 9.8 for men. The highest age specific incidence rate for ulcerative colitis was between 20 and 39 years: 5.5 for women and 6.5 for men. This first French prospective study has shown an incidence rate for Crohn's disease comparable with that seen in north European studies but lower than that seen for ulcerative colitis. These results could be related to the different environmental factors or the genetic background of the population studied, or both.

Inflammatory bowel disease in married couples: 10 cases in Nord Pas de Calais region of France and Liège county of Belgium.

Ten pairs of husband-wife couples are reported with inflammatory bowel disease who were seen in the same geographical area in Nord Pas de Calais region of France and in Liège county (Belgium). Among these 10 couples, four were concordant for Crohn's disease, two for ulcerative colitis, and four were discordant. In nine of 10 couples neither spouse had symptoms before marriage but inflammatory bowel disease subsequently developed in both. In one couple, one spouse had Crohn's disease before marriage and the other partner experienced symptoms afterwards. Eighteen children were born to eight of 10 couples. Five of them developed Crohn's disease but four belong to the same family. In all cases the affected children were born to parents who both developed Crohn's disease after they had married and were conceived at a time when parents did not yet have symptoms. It is proposed that this pattern of emergence of inflammatory bowel disease suggests a role for an infectious agent yet to be identified.

Pulmonary hamartoma and malignancy.

Sixty-five patients, aged 15 to 65 years, including 48 men (73.8%), were operated on for pulmonary (60) or endobronchial (5) hamartoma: there were 36 enucleations, 17 lobectomies (1 sleeve resection), 7 wedge resections, 4 segmentectomies, and 1 pneumonectomy. The average tumor diameter was between 2 and 4 cm; 15 were calcified. All were benign. Two were parts of an incomplete Carney's triad. Eight were associated with carcinomas, and three of these were bronchial carcinomas. Concerning these latter three tumors, a study of the incidence tables for northern France and of the standardized mortality ratio showed that the risk of bronchial cancer developing in patients with hamartoma, after a sufficient follow-up (61 patients), was multiplied by 6.66, and the chi 2 test showed a significant difference of incidence compared with that in the general population (p < 0.001). It is concluded that patients with hamartoma should be submitted to a complete evaluation and to a regular follow-up.

The relationship between the phospholipid fatty acid composition of red blood cells, plasma lipids, and apolipoproteins.

This study examined the relationship between the fatty acid composition of red blood cell phospholipids and lipid markers of atherosclerotic risk in an urban male population aged 45 to 66 years. There was a surprisingly significant positive association between the docosahexaenoic acid ([DHA] 22:6n-3) content of erythrocyte phospholipids and the following risk markers: plasma cholesterol (P < .01), low-density lipoprotein (LDL) cholesterol (P < .01), apolipoprotein (apo) B (P < .05), and apo B-containing lipoprotein particles (P < .05) recognized by a monoclonal antibody (LpBL3). On the other hand, phospholipid alpha-linolenate was positively correlated with apo A-I and high-density lipoprotein (HDL) cholesterol levels (P < .05), while arachidonate showed an inverse relationship with plasma cholesterol (P < .05). There was a negative association between palmitoleic acid and apo B (P < .01) and LpBL3 (P < .001); the latter showed a negative association with stearic acid (P < .001). These interesting findings emphasize the beneficial effect on atherosclerotic risk markers of dietary n-6 polyunsaturated and monounsaturated fatty acids, and suggest that long-chain n-3 polyunsaturated fatty acids (DHA) could have an adverse effect on some of the lipid risk markers.

[What is the prognosis in unclassified colitis? Results of a cohort study of 104 patients in the Northern-Pas-de-Calais region]. / Que deviennent les colites aiguës inclassées? Résultats d'une étude de cohorte de 104 malades dans la région Nord-Pas-de-Calais. Registre des Maladies Inflammatoires du Tube Digestif du Nord-Quest de la France.

Acute unclassified colitis could be the first attack of inflammatory bowel disease, particularly chronic ulcerative colitis or acute non specific colitis regarded as being of infectious origin without recurrence. The aim of this work was to determine the outcome of 104 incidental cases of acute unclassified colitis diagnosed during the year 1988 at a census point made 2.5 to 3 years later and to search for demographic and clinical discriminating data for final diagnosis. Thirteen patients (12.5%) were lost to follow up. Another final diagnosis was made in three other patients: two had salmonellosis and one diverticulosis. Of the remaining 88 patients, 46 (52.3%) relapsed and were subsequently classified as inflammatory bowel disease: 54% ulcerative colitis, 33% Crohn's disease and 13% chronic unclassified colitis. Forty-two (47.7%) did not relapse and were considered to have acute non specific colitis. The mean age at onset was significantly lower in patients with inflammatory bowel disease (32.3 years) than in patients with acute non specific colitis (42.6 years) (P < 0.001). No clinical data (diarrhea, abdominal pain, bloody stool, mucus discharge fever, weight loss) was predictive of the final diagnosis. In this series, 52.3% of patients initially classified as having an acute unclassified colitis had a final diagnosis of inflammatory bowel disease after a 2.5-3 years follow-up. These data warrant a thorough follow up of acute unclassified colitis, especially when it occurs in patients < 40 years.

Increased triglyceride levels in shift workers.

PURPOSE: To assess an independent relationship between shift work and serum lipid levels. DESIGN: Cross-sectional survey. SETTING: Two plants of northern France: a chemical one and a nuclear power station. PARTICIPANTS: All the shift workers of the chemical plant and of one part of the nuclear station. One hundred nine persons were selected, 25 were excluded or absent during the study, and 11 refused to participate. Day workers matched with shift workers according to age, educational level, birthplace, and occupational physical activity level served as controls; 109 were selected, 26 were excluded or absent, and 10 refused to participate. MAIN OUTCOME MEASURES: Fasting venous plasma concentration of total cholesterol, triglyceride, and high-density lipoprotein (HDL) cholesterol; dietary intake assessed by a 3-day record, smoking habits, and body mass index (BMI). RESULTS: Shift workers had significantly higher levels of serum triglyceride (1.26 versus 1.03 mmol/L, p = 0.01). Cholesterol and HDL cholesterol levels were similar for the two groups. There was no difference in energy and nutrient intake, but day workers had a higher alcohol intake (15.64 g/d versus 9.3 g/d, p = 0.03). Multivariate analysis conducted with triglycerides as dependent variable and shift work, BMI, smoking, age, leisure time physical activity level, energy intake, and alcohol intake as independent variables confirmed that shift work has a significant explanatory power for triglyceride levels (beta = 0.134, p = 0.0005). CONCLUSION: This study confirms that shift work is associated with an increase of triglyceride levels independent of dietary intake. We did not find any influence of shift work on cholesterol and HDL cholesterol. Despite this latter fact, our findings are to be considered in the explanation of coronary risk among shift workers. Further studies are needed to elucidate the mechanism of this relative hypertriglyceridemia: stress induced by shift work or diurnal rhythm disturbances.

Relationship between smoking and diet: the MONICA-France project.

The association between smoking habits and diet was examined in a sample of 1126 men, aged 45-64 years, from the general population living in the three French MONICA monitoring areas: Bas-Rhin (BR) (n = 346), Haute-Garonne (HG) (n = 400) and Urban Community of Lille (UCL) (n = 380). Diet was assessed by the 3-day record method. The energy and nutrient intakes were calculated for each of the 3 centres, using the same food composition tables. Alcohol consumption was higher among smokers (P less than 0.001). Taking into account alcohol consumption, age, body mass index, centre, educational level and family size, the analyses showed no difference in non-alcoholic energy intake, proteins, carbohydrates, and total fat. However, smokers had a lower intake of polyunsaturated fatty acids (P less than 0.05) and dietary fibre (P less than 0.01) than non-smokers. With regard to food items, smokers had a higher intake of sucrose (P less than 0.05) and a lower intake of vegetables (P less than 0.001), dairy products (P less than 0.05) and cheese (P less than 0.05). These results suggest that analyses in future epidemiological research regarding the role of diet in the aetiology of tobacco-related diseases should consider this association of potential risk factors.

Relation of arteriographically defined coronary artery disease to serum lipoprotein particles mapped with monoclonal antibodies.

BACKGROUND: This study was designed to investigate the relation of a molecular analysis of apolipoprotein B (apoB)-containing atherogenic lipoprotein particles to coronary artery disease (CAD) in middle-aged men. METHODS AND RESULTS: Two groups of men were studied. The first consisted of 97 patients with angiographically documented CAD (greater than 50% stenosis of at least one coronary artery). The second group consisted of 145 subjects without symptomatic CAD, who served as controls. In both groups, measurements were obtained for total cholesterol level, triglyceride level, cholesterol contents in apoB- and nonapoB-containing particles (LpB, LpnonB), total apoB and apolipoprotein AI (apoAI levels), lipoprotein particles recognized by monoclonal antibodies anti-apoB (LpBL3, LpBL5, LpBL7) and anti-apoAI (LpAI-2GII). Taking into account age, body mass index, hypertension, diabetes, smoking habits, and drug consumption, the analysis showed that the mean levels of cholesterol were identical in both groups but differed when cholesterol content in LpB and LpnonB subfractions were assessed, thus reflecting an increase in the low density fraction and a decrease in the high density fraction, respectively. This was confirmed by an increase in total apoB and a decrease in total apoAI. Measurements of LpBL3, LpBL5, LpBL7, and LpAI-2GII particles also discriminated between the two groups. After adjustment for cholesterol content in LpnonB particles, a difference in total apoB was no longer significant between groups, whereas LpBL3, LpBL5, and LpBL7 levels remained significantly higher in CAD patients. CONCLUSIONS: The measurement of separate concentrations of apoB in different particles may permit a more-accurate assessment of CAD risk than measurements of total apoB levels.

[Pericardial C-reactive protein. A marker of agonal cardiac disease ?]. / La protéine C-réactive péricardique. Un marqueur de la souffrance cardiaque agonique?

We studied the influence of the agonal period on the concentrations of acute phase proteins in biological fluids obtained from 26 autopsy cases. We found significant differences for C-reactive protein concentrations in serum and in pericardial fluid, between short and long agonies. The other acute phase proteins studied (alpha-1 antitrypsin, alpha-2 macroglobulin, haptoglobin) failed to show any significant difference in serum and pericardial fluid levels between the two types of agony. The increase in C-reactive protein level in the pericardial fluid is attributed to an "agonal pericarditis" which may result from an agonal myocardial necrosis. Our results could be of interest in forensic medicine.

[Primary results of the MONICA project]. / Les premiers enseignements du projet MONICA.

The MONICA project is an international study coordinated by the WHO, designed to explain the important variations of coronary mortality observed in the industrialised countries over the last 20 years. Thirty-nine centres in 27 countries are participating in the project which, over a 10 year period will, in geographically determined zones; a) record the numbers of acute myocardial infarcts and coronary deaths, b) analyse the treatment of acute cardiac events, and c) evaluate the cardiovascular risk factors and preventive measures in the general population. The MONICA-France project comprises three registers (Bas-Rhin, Haute-Garonne and the urban community of Lille), and a coordinating centre. The preliminary results confirm the wide geographic variability of coronary mortality with a prevalence slightly higher in Alsace and the North than in the South-West of France. The approximative frequency of coronary events in France is estimated at 112,000 infarcts and over 175,000 acute coronary episodes and deaths. Information is provided about the severity of the principal coronary risk factors and the conditions of their treatment in the register zones. A study of the accessibility of emergency treatment of acute infarction in the Bas-Rhin area, has demonstrated the key role of the general practitioner as the first contact but confirmed the long delay to hospital admission which nullified the potential benefits of thrombolysis in a high proportion of cases. However, the significant decrease in hospital mortality of acute infarction observed over a 3 year period in the three register zones, is probably related to the number of patients thrombolysed in that time.

[Validity of death certificates in the study of ischemic heart diseases]. / Validité des certificats de décès pour l'étude des cardiopathies ischémiques.

Mortality rates for ischaemic heart disease, as estimated from death certificates, show highly significant differences between countries. In order to study the validity of mortality rates for ischaemic heart disease, the authors, involved in the MONICA project, have compared the results obtained from the conventional death certificate code with the data collected in a complementary enquiry conducted for all deaths possibly due to ischaemic heart disease. Three hundred and thirty patients, aged from 25 to 64 years, belonging to the urban community of Lille, and who died between October 1 and December 31, 1984, were included in this study. The sensitivity of the death certificate for the diagnosis of ischaemic heart disease was 77.9 percent and its specificity was 95.9 percent. The concordance rate between death certificate and complementary enquiry was not modified by age, sex, socio-professional category, family situation, place of death and doctor who signed the certificate. The complementary study proved impossible in 31.8 percent of the cases, usually because the doctor who signed the death certificate was not fully conversant with the patient's condition. Our results therefore confirm that death certificates are valid to study mortality from ischaemic heart disease.