Environmental data
Water outage data for the study period were provided by the Taiwan Water Corporation. These data were used to identify affected areas, dates of occurrences, duration, and reasons for outages, such as broken pipes or constructions.
Considering the effects of weather on infectious diseases, we used daily weather measurements and events data from the Central Weather Bureau (CWB). In the present study, typhoons were defined based on typhoon warnings released by the CWB, which contained information on time, duration, and areas hit by each typhoon that appeared during the summer months. We also used daily average temperatures to evaluate the relationship between infections and temperatures.
Aside from typhoon data, flood data from the National Fire Agency, as well as data on affected areas, dates of occurrences, and receding dates from the Ministry of Interior Affairs, were also used.
To clarify further the association between contamination of drinking water and water outage, we also obtained drinking water quality inspection data from Taiwan's Environmental Protection Administration's (EPA) Executive Yuan for data analyses. Data on the most probable number of coliform tests in water samples were used.
Health insurance data
These insurance claims were randomly selected from the National Health Insurance database, which was established and provided by the National Health Research Institute (NHRI). The reported coverage rate among the population of 23 million has been 96% or higher since 1996 [7]. The medical claims data provided information on scrambled identifications, gender, birthdays, types of services and diagnoses, dates of admissions and discharges, and medical institutions providing services for each patient. Eight cities and counties were chosen to represent urban and suburban areas for northern (Taipei and Taoyuan), central (Taichung and Changhua), southern (Kaohsiung and Tainan), and eastern (Taitung and Hualien) Taiwan. The present study used medical claims of one million insured persons and environmental data for 2004-2006. The total study population size covered in the insurance system in these areas in 2005 was 624,176. The health insurance data used in this study obtained from the NHRI have been safeguarded for the privacy and confidentiality with scrambled identifications. This study was thus exempted from ethical review.
We retrieved medical records before, during, and after each water outage for medical services for gastroenteritis, eye and skin infections, and other complaints, in accordance with the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9 CM). The usage of medical services in the present study includes outpatient visits, and emergency and inpatient cares combined (EICC). The ICD-9-CM codes representing gastroenteritis were ICD-9 CM 001-009, 535, 5362, 555, 5582, 5589, 567, 5689, 578, 787, and 789. Infectious skin diseases (ICD-9 CM 680-686), acariasis (ICD-9 CM 133), mycoses of skin (ICD-9 CM 110 and 111), and rashes (ICD-9 CM 7821) were selected from a group of skin diseases. Meanwhile, codes for eye diseases were conjunctivitis (ICD-9 CM 3720), inflammation of the eyelids (ICD-9 CM 3734, 3735, and 3736), and trachoma (ICD-9 CM 076).
Data analysis
Considering that the effects of water outages could be delayed for a few days, we divided each water outage event into three periods, namely, 10 days with normal water supply before any water outage (normal period), the actual day(s) with water outages ( lag 0), and 10 days after the outages (10-day lag). To exclude effects from typhoons and floods, water outages during these weather events were excluded.
We calculated the incidence rates of medical services, including outpatient visits and EICC for gastroenteritis and eye and skin complaints in person-days in normal, lag 0, and 10-day lag periods. Both the outage period rate to the normal water supply period rate ratio and the 10-day lag period rate to the normal water supply period rate ratio were calculated separately for each type of infection.
We performed the Poisson regression model to measure the risks of medical visits associated with water outages. Afterward, we estimated the average daily temperature (<15, 15-20, 20-24, 25-29, and 30+°C) specific outage-to-normal relative risks (RR), and 95% confidence interval (CI) for selected diseases. Adjustment was made for calendar year, month, holiday, geographic area, sex, age, gross domestic product (GDP) index, and education index. The multivariate-adjusted models, which served as a control for the daily average temperature instead of age (<15, 15-64, and 65+ years), were repeated to estimate the age-specific outage to normal-period RR. The model is specified as follows:
Log (∆Yi) = α + β1X1 + ... + βpXp.
RRs of daily area-disease-specific outpatient visits and EICC (ΔYi) associated with water outage period and a 10-day lag period, compared with the normal period, were estimated after controlling for the covariates, such as area, sex, age (<15, 1-64, and 65+ years), daily average temperature (<15, 15-20, 20-24, 25-29, and 30+°C), daily relative humidity, GDP index, education index, year, and month. RR and 95% CI were calculated based on the exponential transformation of βi estimations.
To clarify whether water outage-related diseases resulted from water-borne or water-washed pathways, we compared the water supply quality during the three observed periods of the water outages. Based on test results, Taiwan EPA recorded each tested drinking water sample as either "unqualified" or "qualified." Matching with the sampling time, we calculated the period-specific "water unqualified rate (%)" by dividing the number of unqualified water samples by the total number of tested water samples. The Chi-square test was used to check if the water test frequency differs with periods. The association between unqualified rate (%) and water outage period (normal period, lag 0, and 10-days lag) was evaluated using generalized linear models. All statistical analyses were performed using SAS version 9.1 (SAS Institute Inc., Cary, NC, USA).