Annual Incidence Rate of Infectious Diseases Estimated from Sentinel Surveillance Data in Japan

BACKGROUND: The estimation of incidence rates of infectious diseases based on the sentinel surveillance data is rather rare. We attempted to estimate these in 2000 in Japan by the surveillance data, and to evaluate their biases. METHODS: We used the incidences of influenza-like illness and 12 pediatric diseases in each of the sentinel medical institutions in Japan based on surveillance data in 2000. The incidence in all medical institutions was estimated under the assumption that the sentinel medical institutions were randomly selected. The possible bias of this estimate was evaluated in comparison with the hypothetical true incidence obtained as the total incidence in all medical institutions estimated by a regression model using the numbers of all disease outpatients per day from the National Survey of Medical Care Institutions of Japan. RESULTS: The estimated annual incidence rate was 75.6 (95% confidence interval: 72.3-78.7) per 1,000 population in influenza-like illness, and ranged from 1.1 (95% confidence interval: 1.0-1.2) to 285.2 (95% confidence interval: 270.2-300.3) per 1,000 population aged 0-19 years among 12 pediatric diseases. The ratio of the estimated incidence to the hypothetical true one was 1.06-1.26 among influenza-like illness and the 12 pediatric diseases. CONCLUSIONS: The incidence rates of influenza-like illness and pediatric diseases in 2000 in Japan were estimated from sentinel surveillance data. The rates obtained provide some useful but not always accurate information. Thus, further research is necessary.

The surveillance of infectious diseases has been established in many countries.1-10The estimation of the incidence rates of infectious diseases based on the sentinel surveillance data is rather rare, although such surveillance provides some useful information regarding incidence.11-13In many surveillance projects including one in Japan, sentinel medical institutions (SMIs) are recruited on a voluntary basis.However, uncertainties remain as to how representative the finding are when applied to all the medical institutions in given areas.In the method for estimating incidence rates by surveillance data, it would be assumed that SMIs are randomly selected from all the medical institutions if information on the underlying SMI population is not available.11,14 In Japan, the guidelines for the surveillance, introduced in 1999 by the Ministry of Health, Labor and Welfare, determine that SMIs are selected from all medical institutions in the areas as randomly and as representatively as possible (http://idsc.nih.go.jp/index.html).15 Prefectural governments select SMIs according to the guidelines provided.The numbers of all disease outpatients per day in each medical institution from the National Survey of Medical Care Institutions conducted by the Ministry16,17 are available and would be useful to evaluate the underlying SMI population.We attempted to estimate the incidence rates of infectious diseases in 2000 in Japan from surveillance data under the assumption that SMIs were randomly selected from all medical institutions.We also sought to evaluate the possible bias in these estimated incidence rates using the number of all disease outpatients per day in each medical institution.

Surveillance of infectious diseases in Japan
The Surveillance data and method for estimating incidence rates Table I shows the numbers of all medical institutions and SMIs in 2000.About 94% of SMIs reported all 52 weeks in 2000, and the others started or stopped reporting in 2000.The pediatric disease surveillance system included 3,011 SMIs, which were departments of pediatrics in hospitals and clinics with a department of pediatrics.The influenza-like illness surveillance system included 4,656 SMIs, which were departments of internal medicine or pediatrics in hospitals and clinics with departments of internal medicine and/or pediatrics.Those included most SMIs for the pediatric disease surveillance system.Proportions of SMIs among all medical institutions ranged from 1.3% in clinics with a department of internal medicine and without a department of pediatrics, to 30% or more in clinics with a department of pediatrics and without a department of internal medicine.The numbers of all medical institutions were obtained from the National Survey of Medical Care Institutions conducted by the Ministry in October 1999.16 We used 238,201 reports from the influenza-like illness surveillance system and 155,184 reports from the pediatric disease surveillance system.These reports covered about 98% observations over 52 weeks in each of the SMIs in 2000.Each report included the sex-and age-specific numbers of influenza-like illness patients newly diagnosed during a given week in the influenza-like illness surveillance system and the numbers of patients aged 0-19 years with 12 pediatric diseases (see Table 3) in the pediatric disease surveillance system.In each SMI, the sex-and age-specific annual incidences of influenza-like illness and the annual incidences of each pediatric disease in the patients aged 0-19 years were calculated as the totals of those in weekly reports.
The missing data (less than 2% among the total reports) were replaced with the mean incidence calculated for each disease, week, prefecture and type of medical institution (specifically also for each sex and age group in influenza-like illness).The medical institutions for pediatric diseases were classified into 3 types.
departments of pediatrics in hospitals, clinics with only a department of pediatrics and clinics with departments of internal medicine and pediatrics.Those for influenza-like illness were classified into four types, the above three types and others, because the number of SMIs in either the departments of internal medicine in hospitals or clinics with only a department of internal medicine was 1 or less in several prefectures.
For each disease, prefecture, and type of medical institution, the incidences in SMIs follow a multi-hypergeometric distribution under the fixed condition of the total number of SMIs under the assumption that SMIs are randomly selected from all medical institutions.The total incidence in each prefecture and type of medical institution was estimated as the total incidence in SMIs Annual Incidence of Infectious Diseases divided by the proportion of SMIs to all medical institutions.The total incidence in all medical institutions was estimated to be the total of those in all prefectures and types of medical institution.
The approximate confidence interval for the incidence was given based on the distribution.The appendix shows the method for estimating incidences in detail.
Method for evaluating biases of estimated incidence rates We used the numbers of all disease outpatients in September 1999 in each department of internal medicine and pediatrics in hospitals and those over one week in September 1999 in each clinic from the National Survey of Medical Care Institutions of Japan.16 The number of all disease outpatients per day was used in 4,585 out of 4,656 SMIs and in 76,412 of all medical institutions for estimating the incidence of influenza-like illness, and in 2,991 out of 3,011 SMIs and 30,828 of all medical institutions for estimating the incidences of pediatric diseases.
For each disease and type of medical institution, a linear regression model with a dependent variable for the incidence and independent variables for the number of all disease outpatients per day and prefecture (as dummy variables) was applied using the SMI data.The hypothetical incidence in every medical institution was calculated as the expected value using an estimated linear regression equation and the data of its independent variables.
Using the hypothetical incidences in all SMIs, the incidence in all medical institutions was estimated by the above-explained method.The bias of this estimate was evaluated in comparison with the hypothetical true incidence obtained as the total hypothetical incidence in all medical institutions.

RESULTS
Table 2 shows the estimated incidence of influenza-like illness in 2000 in Japan.The annual incidence was estimated to be 9,590,000.The annual incidence rate per 1,000 population was estimated to be 75.6 (95% confidence interval [CI]: 72.3-78.7).The rate was slightly higher in males than in females, and decreased with age.
Table 4 shows the numbers of all disease outpatients per day in all medical institutions and SMIs.For each type of medical institution, the mean number of all disease outpatients per day was higher in SMIs than in all medical institutions.
Table 5 shows the incidences of influenza-like illness and 12 pediatric diseases estimated from the hypothetical incidences in all SMIs, and their hypothetical true ones.The ratio of the estimate to the hypothetical true one was 1.00-1.09among influenzalike illness and 2 pediatric diseases, 1.10-1.19among 8 pediatric diseases, and 1.20-1.29 among 2 pediatric diseases (pertussis and measles).
Table 2.Estimated incidences of influenza-like illness.
The incidence rate and its 95% confidence interval are stated per 1,000 population.
The incidence rate and its 95% confidence interval are stated per 1,000 population aged 0-19 years.
Table 4.The numbers of all disease outpatients per day in all and sentinel medical institutions.
SD: standard deviation Table 5.Estimated and hypothetical true incidences of influenza-like illness and pediatric diseases.
The estimated incidence was obtained from hypothetical incidences in all SMIs.
Annual Incidence of Infectious Diseases

DISCUSSION
The incidence rates of influenza-like illness and pediatric diseases in 2000 in Japan were estimated from surveillance data under the assumption that SMIs were randomly selected from all medical institutions.No information strictly comparable with our results was available because the incidence rates of infectious diseases varied widely among years and areas.Compared among the weekly numbers of patients per SMI in 1990-2000 in Japan (http://idsc.nih.go.jp/index.html),the epidemic in 2000 was large in group A streptococcal pharyngitis, pertussis and herpangina, small in rubella, and moderate in influenza-like illness, chickenpox and other diseases.The estimated incidence rate of influenzalike illness in 2000 in Japan (75.6 per 1,000 population) was higher than the figures reported in United Kingdom and France (16 and 42 per 1,000 population in the epidemic period of 1993 and 1995/1996, respectively), 14,18 and within the range reported in The Netherlands (15-86 per 1,000 population between 1971-1989). 3 The finding in this study that the incidence rate of influenza-like illness decreased with age, has been already reported.14,19The incidence rate of chickenpox, as an example of pediatric diseases, is to be compared with those among several countries.The estimate in this study (90.3 per 1,000 population aged 0-19 years) was converted into about 18 per 1,000 population of all ages because the incidence rate of chickenpox was very low in population aged 20 years or over.This figure was higher than reported in UK and France (2-9 and 10-13 per 1,000 population in 1967-1985 and 1991-1995, respectively).20,21 The assumption of random selection of SMIs was critical.11We attempted to evaluate possible bias in the estimated incidence using the number of all disease outpatients per day.The mean number of all disease outpatients per day was higher in SMIs than in all medical institutions.Although the number of all disease outpatients per day would not be the best index to the underlying SMI population, especially in infectious diseases with a great seasonal variation such as influenza-like illness, it suggested that the mean size of the underlying population was larger in SMIs than in all medical institutions, and that the assumption of random selection of SMIs would not be strictly valid.Since this assumption was undermined, the incidence rates of infectious diseases were overestimated.The ratio of the estimated incidence to the hypothetical true one was calculated to be 1.06-1.26among infectious diseases and to be over 1.20 in pertussis and measles.However, the ratio would not be sufficiently accurate to be used for adjusting the estimated incidence rate.In the infectious diseases in which the ratio was less than 1.1 or 1.2, the estimated incidence rate might be useful for public health activities.The estimated incidence rate and its ratio to the hypothetical true one, even though not strictly accurate, provide some helpful information for planning vaccinations against infectious diseases such as influenza and rubella.4,19 In this study, we encountered several further problems and limitations in estimating the incidence rate from sentinel surveillance data.Although the criteria for diagnosis of infectious diseases were determined in the guidelines by the Ministry of Health, Labor and Welfare of Japan,9,15 the reports from SMIs might not always be based on such criteria.The incidence estimated in this study did not include persons with infectious diseases who had not visited medical institutions because information was available only from medical institutions.This limitation should be kept in mind when regarding the figures.Prefectures and types of medical institution were used as the strata in the method for estimating incidence rates.The assumption in the method was that SMIs were randomly selected in each stratum, not in all medical institutions.The strata used would be only natural because prefectures selected SMIs.The underlying population and patient characteristics such as age would differ greatly between hospitals and clinics and among departments as well.As described above, departments of internal medicine in hospitals and clinics with only a department of internal medicine were combined as to the type of medical institution for estimating the incidence rate of influenza-like illness because the number of SMIs in either the departments of internal medicine in hospitals or the clinics with only a department of internal medicine was I or less in several prefectures.
We think that the incidence rates of infectious diseases would not be completely accurate when estimated by only the information obtained from routine sentinel surveillance.However, these estimates, albeit somewhat biased, would still be useful for public health activities.Further research of methods for estimating incidence rates from sentinel surveillance data and evaluating their biases is important.
is organized by the Ministry of Health, Labor and Welfare.9,10It involves the systems for influenza-like illness and pediatric diseases.Local governments select SMIs according to the Ministry guidelines.The numbers of SMIs in the areas covered by public health centers are approximately proportional to their population sizes.In reality, SMIs seem to be recruited on a voluntary basis to some extent, although the Ministry guidelines call for the SMIs to be selected from all the medical institutions in the areas as randomly and as representatively as possible.15Each SMI reports the numbers of cases of notifiable infectious diseases to the area public health center weekly.Notification by public health centers to the local government and the Ministry is made through an on-line computer network.

Table 1 .
The numbers of all and sentinel medical institutions.
The percentages are in relation to the number of all medical institutions.