概述：

很少有人知道病人之間的相關性病癥。然而，衛生保健管理部門應該對病人的癥狀調查所持有的大量信息感興趣。我們提出了一種新的技術基于考慮癥狀的競爭評分，來解釋他們的在病人之間的流行程度。

來自貝寧的數據，我們判斷的多項對分數模型對四種癥狀相對的患病率，為患病者的四個年齡段的病人，年齡和性別，婚姻狀況，家庭組成，衛生設備和有毒的消耗性疾病影響著相關的癥狀流行。此外，生活質量和經濟活動也扮演著重要的角色，顯示窮人和從事農業的病人的患病癥狀不同于富有的人或普通人。對衛生保健管理三種結果類型可以從判斷的結果提取。首先，由于很少人知道社會人口和經濟變量在患病人群中影響著癥狀的模式，任何新的相關性的系列描述是受歡迎的。第二，這樣的判斷結果，在因果關系的特殊研究協會可以針對特定的癥狀或特定群體用于協助衛生保健管理。第三，這個判斷可能揭示變量，可以作為一個健康的干預工具。，

Abstract

ittle is known about the correlates of symptoms among ill persons. However, surveys of symptoms among ill persons hold much information that should be of interest for health care management. We propose a new technique based on the consideration of competing scores of symptoms to explain their relative prevalence among ill persons.

Using data from Benin, we estimate multinomial logit models of relative prevalence of four categories of symptoms,for four age classes of ill persons. Age and gender of the ill person, marital status, household composition, health equipment and noxious consumption influence the relative prevalence of symptoms. Moreover, living standards and economic activities also play important roles showing that the pattern of symptoms of poor and agricultural ill persons is different from that of the rich or of the non-peasants. Three types of outcomes for management of health care can be extracted from the estimation results. First, since very little is known about the socio-demographic and economic variables affecting the structure of symptoms among ill persons,
any new descriptive series of correlates is welcome. Second, such estimation results, in association with specific studies of causal relationships can be used to assist health care management directed towards specific symptoms or specific groups. Third, the estimation may reveal variables that might be used as instruments for a health intervention.

1 The Model模式
We propose an approach based on the assumption that every observed symptom is characterised by its “latent potentiality” (or score) due to the environmental factors and the host characteristics associated with these explanations. Indeed, as suggested Mausner and Bahn (1985), or Lilienfeld and Stolley (1994), the development of a specific illness must account for successive stages: susceptibility; presymptomatic phase; clinical disease; and perhaps disability or death. During the subclinical stages, no symptom can be observed. Thus, symptom scores characterise the level of development of diseases. The score of the ith disease (Si) is an indicator of its latent severity and describes the "spectrum of the disease"4. The first part of this spectrum (susceptibility or presymptomatic level) corresponds to unobserved events occurring in the human organism from the time of exposure. Although unapparent, these events associated with early infection are important because they play a role in the transmission of infectious agents. When the disease is actually observed, the score corresponds to the level of clinical symptoms. We assume that the declared symptom corresponds to the highest score5 obtained among the set of symptoms.#p#分頁標題#e#

2 The system of equations方程組
Our approach is original on several grounds. First, we focus on the population of ill persons that is the relevant target of health intervention, and the type of population that is likely to be encountered at treatment sites (hospital, community health center). Second, we do not model the incidence of diseases but rather their relative prevalence, which is consistent with the fact that we observe a sample of ill persons. Third, we consider competing symptoms instead of isolated diseases. Indeed, for populations affected by multiple health problems, concentrating the heed on an unique disease may give a misleading picture of a complex situation.

We include in X, available variables that describe mostly the host factors. These variables are socio-demographic characteristics of the ill person and of the household; variables related to economic status or behaviour likely to influence health status; and health equipment. Unfortunately, the environment factors are unobserved. We control for their influence by
including dummy variables for seven districts.

3. Data Description and Econometric Results數據描述和劑量結果
3.1. The data
3.1.1. The country
Benin is a small rural country in western Africa with a population of 4.74 millions in 1991, with 48 percent under 15 years old10. Per capita GNP is US $ 340 which makes Bénin one of the poorest country in the World. Agriculture is the cornerstone of Benin's economy and contributes 37 percent of the country’s GNP. Moreover, three-fourths of the active population is employed in the agricultural sector. The education level of the population is very low with around four-fifth
of adult people illiterate. Health status is also dramatically low. Average life expectancy at birth is estimated at 50 years. Children experience a heavy mortality toll with perinatal mortality rate equal to 6.9 percent and juvenile mortality rate equal to 17. This high mortality is partly due to morbidity, especially from numerous endemic diseases such as malaria, parasitises and tuberculosis. Malnutrition is also widespread with 35 percent prevalence of malnutrition amongst children under five years old, whereas 10 percent of children show weight insufficiency at birth.
Despite the extent of poverty in the country, efforts have been made to improve the population’s health. 67 percent of children are vaccinated with their third dose of DCT and 34 percent of deliveries benefit from an health assistance. But households spend on average only 5 percent of their final consumption expenditure on health, whereas 37 percent goes to food consumption. Moreover, only 4.3 percent of the GNP is devoted to health expenses with 41.8 percent of these expenses being funded by international assistance. The knowledge of factors associated with the main type of diseases may help to enhance the efficiency of the too scarce resources allocated to health.#p#分頁標題#e#

3.1.2. The ill person sample病人的樣本
The data is taken from a random health survey conducted by the government of Bénin covering the district of Ouidah, which has about 70 000 inhabitants, in the South-East of Cotonou, from May to September 1992. This district is composed of nine communes where 2591 households were visited, corresponding to 11502 individuals. 880 individuals reported having suffered an illness at the period of the collection11. Due to missing values, 786 observations of ill persons are used in the estimation12. The household sample is representative of the district of Ouidah. The household members who reported an illness or a disease two weeks prior to the interview have been asked about their health and their socio-demographic and economic characteristics at the individual/household level. The interviews were conducted by doctors and medical staff from the Community Health Centre in Pahou. The health knowledge of these enumerators suggests that the measurement of health status is more precise than usual13. The basis of enquiries is a mixture of self-report and medical examination. Diagnostic rather than disease has been recorded to avoid misreporting from individuals who self-treated.

 We do not dispose of the data for the whole sample of households but only for the sample of ill persons in households with illness. However, we know that the whole sample of households is characterised by a smaller average household size (4.40 members and a lower average education level (64.7 percent of uneducated people to be compared with 36.5 percent for the ill persons). Finally, the proportion of females is similar in the two samples (52.5 percent for the whole set of
households, 55.73 for the ill persons). These differences between ill person households and non ill households illustrate the selectivity associated with the constitution of the ill person population. However, because our concern in this paper is the study of competing illnesses for ill persons, and not the incidence of illness in general, we do not incorporate this selectivity mechanism in the model, which would be anyhow impossible to identify from the available information. We divide our sample of ill persons in four age classes so as to account for the specificity of health processes for different stages in life-cycle (Mausner and Kramer (1985)). Indeed, age is related to the occurrence of infectious diseases and to their severity, and chronic diseases tend to increase with age. 171 ill persons are babies (under four years old); 132 are young children(between 4 and 10); 71 are adolescents (between 11 and 18); and 412 are adults (over 18).

Table 1 shows the mean and standard deviation of the main variables by age class of the ill persons.
Let us consider first the whole sample of ill persons. The average age of ill persons is about 27. The average household size is of 5.08 members. Amongst them around two-fifth (38 percent) are ill. Most of these members are children (4.25 by household) and the average age of ill persons is 27.3 years. Only one-fourth of ill persons are educated with an average education level of 0.44 years (including children too young to be educated). Thirty-six percent of households are headed by a female head. The average age of the household head is 44.4 years. About 40 percent of ill persons live in a peasant household but only 7 percent of households are totally dependent on agricultural activity. On average 5.39 active persons from the family assist in household activities, mostly for agricultural work. Twenty-seven percent of households in which lives the ill person, are corn producers. A majority of ill persons use kerosene for lighting (83 percent), thus avoiding the noxious effects of smoke from firewood, although only 39 percent are equiped with septic tank, showing a general susceptibility to contagion through excrements.#p#分頁標題#e#

Declared total expenses amount to 10475 FCFA 14 on average, which corresponds to average per capita expenses of 3166 FCFA. Only 1614 FCFA have been spent on health expenses in those households where at least one member is ill. This amount is to compare with the level of average tobacco expenses (73 FCFA) and the level of average alcohol expenses (444 FCFA)15 that are believed to be noxious to health status. Communes 1, 6 and 8 have been combined together because their individual sample size was not sufficient for statistical analysis. The location of the commune in the urban community of Ouidah (43 percent of the sample), or the average distance to modern health centres are controlled by dummy variables.

We examine now the characteristics of ill persons in specific age classes. The demographic and economic characteristics of households do not vary very much with the age class of the illperson16. Naturally, age and education level are increasing in the age class. Half of ill persons are female for age group 0-3 years (babies) or 4-10 years (children), and respectively 58 and 60 percent for age group 11-18 (adolescents) and 19+ (adults). Some symptoms are more frequent for specific age classes: cough, diarrhoea, skin diseases for babies; fever for children; fever, wounds, abdominal pains for adolescents; abdominal pains, fatigue, articular pains, cardiovascular illnesses for adults.


3.1.3. The symptoms癥狀
The average duration of illness is 10.85 days, although since this variable has been truncated to 15 days for the longest duration, the mean underestimates the actual mean duration. Illnesses and diseases are recorded in terms of symptoms, which are classified in to 33 categories.

Table 2 shows the frequency of occurrence for every symptom. Most of the usual health problems
are observed. Only three symptoms show a sufficient number of observations for econometric analysis: "Fever", "Cough" and "Wounds". They represent 61 percent of observations and we focus on their relative prevalence. We group all the other symptoms in a residual category "Others". The diseases are classified on the basis of symptomatology rather than on etiology which would have necessitated the knowledge of the specific agent of each illness17. Surprisingly
diarrhoea, dysentery, malnutrition and other diseases related to diet do not seem to have been correctly recorded, probably because the survey methodology was better adapted to the study of illnesses than to nutrition problems. A nutritional survey based on anthropometric measures would
have been useful for the knowledge of nutritional status. "Fever" accounts for almost half of the symptom declarations. Fever (without cough) may often be attributed to malaria that is frequent in this area18. The symptom "Cough" is moredifficult to attribute to specific diseases. It may be resulting from dengue, influenza, tuberculosis, infantile illnesses, etc. "Wounds" may also have various origins related to working activities, to presence where violent behaviour is happening, to engaging in hazardous activities.#p#分頁標題#e#

Demographic variables are often considered as correlated with morbidity. Gender is believed to be influential and if death rates are higher for males than for females, morbidity rates are generally higher for females. The family size is associated with complex effects since it may be related to living standards (in LDCs rich households have sometimes more children) or to poverty since in large families many persons have to share limited resources. Moreover, the existence of large families helps contagion, although adults may have better health experience and awareness since they had opportunities to practice on children.

In fact, the characteristics of the ill person, and of the household she/he belongs to, vary with the recorded symptom. Table 3 shows these characteristics for the four categories of symptoms. Age, as we have shown above, gender and education of ill persons are linked to the recorded symptoms. Cough and Fever affect relatively less frequently the household head, while Wounds affect rarely the spouse of the head. The average duration of illness is shorter for fever and cough than for wounds and other diseases.

The characteristics of households in which each type of symptoms occurs are also of interest. Wounds are more common in household with larger size, larger number of children or larger number of active persons. On average, Fever is more often associated with a greater land area and with male heads. Cough is more frequently associated with landlords, bachelor head or older heads, absence of septic tank, small land area and production of corn. Although the total expenses and the per capita expenses do not vary much with the symptom, health expenses (lower for Cough), alcohol expenses (higher for Wounds), tobacco expenses (higher for Fever and Wounds) are much more related with observed symptoms. All these associations suggest the possibility of discrimination between the relative prevalence of different symptoms. In the next section, we present estimates of a multinomial logit model describing this discrimination. This enables us to account for the multivariate interaction of correlates, which is impossible with simple descriptive statistics that may provide a misleading picture of effects of variables of interest on ,the prevalence of symptoms.

3.2 Econometric Results劑量經濟學的結果
Table 4 shows the multinomial logit estimates. The interpretation of parameters is in terms of relative effects of the considered variable on the score of the considered symptom relatively to the score of the reference set of symptoms. Therefore, a positive parameter indicates a positive effect on the relative prevalence of the considered symptom (Fever, Cough, Wounds) with respect to the prevalence of symptom “Others”19. Dummies for communes have been included to account for health environment of variable quality20, although the corresponding coefficients are not shown in the table. Many econometric studies deal with the incidence of general illness, or with the incidence of a specific disease or symptom in a particular population. We present results of relative prevalence of symptoms in a population of ill persons21. The age of the ill person is the main demographic variable, consistently with the medicine, epidemiological and health economics literatures, in which health processes are considered strongly dependent on age groups (Souhami and Moxham (1994)). Age should as well affects the composition of diseases in a population of ill persons. For example, the risk of infantile and other illnesses is often strongly decreasing with the growth of children (see the case of Rhinovirus infectious diseases in Rampey et al. (1992); diarrhoea and illness of children in Heller and Drake (1979); diarrhoea and respiratory diseases of infants in Cebu Study Team (1992)). The differences in estimates for each age class and for the global sample supports the use of age classes for the study of relative prevalence of symptoms. Furthermore, the effects of age is significant inside eachage class at least for one symptom, suggesting that age classes of smaller width would be appropriate, although this is not possible with our sample size.#p#分頁標題#e#

Significant effects of age inside age classes show that when they grow older. Babies are more affected by fever and cough, which may be related to the high occurrence of infantile diseases as soon as babies cease to benefit from the immunity provided by the mother’s milk. Young children suffer relatively more from wounds, perhaps because of a greater autonomy of movement at older ages. Adolescents suffer relatively less from cough and adults relatively less from fever. The effect of gender is less strong. However, as expected from absolute incidence studies (Mausner and Kramer (1985)), females are associated with a lower relative prevalence of wounds that males. In particular, wounds are significantly less relatively frequent among female adolescents than among male adolescents. The effects of the education of the ill person is not significant and has been omitted as well as other minor insignificant variables. No information about parent’s education is available.

In absolute incidence studies, the marital status has been found associated with health status. Sickles and Taubman (1997) show that marriage has a positive effect on life-tables. Using Kenyan data, Gage (1997) shows that children of married mothers have a lower probability of polio dropout and of acute malnutrition. Mausner and Kramer (1985) insist that marital status is associated with lower level of mortality for both sexes. Several explanations have been proposed for the influence of the marital status of the health person on her/his health status. Single persons lead often a more dissolute lifestyle. Marriage brings positive interactions between spouses (caring and companionship), although also stress and anxiety. Surviving widows or widowers experience acute grief provoking health problems and premature death. Finally, marriage is in itself a selection mechanism likely to often exclude persons with chronic illness. For adults in our sample, the spouse of the head is a woman. These wives experience relatively less frequently Fever and Wounds. Ill persons in households with single heads have relatively less often Cough. The signs of the coefficient of the two variables describing marital status show that ill persons with symptom in the category “Others” are relatively more frequent for the head’s spouse or in households with
single head. Anyhow, the marital status is not neutral for the relative prevalence of symptoms of ill persons.It is well known that crowding people indoors reinforces the occurrence of respiratory diseases (Sutton (1981)). This implies that, at constant dwelling area, large size households should have a higher incidence of these diseases and the same is true for most infectious illnesses. This is clearly not the only influence channel of household composition since large household size is often correlated in LDCs with high income or socio-economic status, but also with considerable economic burden. In that case, high income might induce both large household and large dwelling, and the effect of household size may not be interpretable in terms of density. Moreover, subtle health management decisions may incorporate the birth order, the number of children or the masculinity ratio of household, directing care and resources allotted to infants and young children (Mausner and Kramer (1985), Strauss and Thomas (1995)). The number or the proportion of active members is an indicator of the capacity of the household to meet its members needs both financially and in terms of caring time. We find here that a large household size lowers the relative frequency of fever among adolescents. Moreover, the number of active persons negatively influences the relative frequency of fever for children and adults, and of cough for children. The structure of disease prevalence of ill persons is clearly dependent on the composition of the household they belong to.#p#分頁標題#e#