Data sources
A universal health insurance program has started in Taiwan since 1995 [4] and covered nearly all inhabitants (21,869,478 beneficiaries at the end of 2002, equivalent to a coverage rate of 97.1%) [5]. The NHI offers extensive hospitalization, ambulatory care and drugs with low co-payment rates. The ambulatory services of Western medicine, dentistry and traditional Chinese medicine belong to independent physicians' clinics and outpatient departments of hospitals. Although a specialty of family medicine does exist in Taiwan, the NHI beneficiaries are not required to register at a general practitioner as in British National Health Service system. The NHI does not establish a referral system of Western standard. The people have great freedom in choosing healthcare facilities and specialties. The outpatient departments, even at the academic medical centers of the tertiary care, are accessible at any time. The reimbursement within Taiwan's NHI is basically on a fee-for-service (fee for each item of services) basis under the global budgeting (similar to the German system). The hospitals and the NHI don't set the limit of visits within a day for each patient.
In 1999, the Bureau of National Health Insurance began to release all claims data in electronic form to the public under the project of National Health Insurance Research Database (NHIRD) [6]. Since then, dozens of extracted datasets has been available to researchers. The structure of the claims files had been described in details on the NHIRD web site and in other published works [7–9].
Every resident in Taiwan has a unique identification number officially and every healthcare facility in contract with the NHI has a unique identification number for claims purpose. The identification numbers of persons and healthcare facilities in the datasets have been encrypted to protect privacy. The encryption is consistent across all datasets so that the encrypted identification numbers remain unique, making longitudinal follow-ups feasible.
One special kind of datasets is the "cohort datasets" which include the claims data of 200,000 persons from 1996 to 2004 (status: November 2005). These 200,000 persons were randomly sampled from 23,753,407 persons who had ever been insured under the NHI from March 1, 1995 to December 31, 2000. The purpose of cohort datasets is to follow up a representative group of the population longitudinally. New claims data of the cohort would be released every year. Not every person of the cohort was insured throughout the whole period because of death and emigration. Besides, those who enrolled in the NHI at the first time after January 1, 2001 would not be included in the cohort.
According to the NHIRD, the randomization in sampling the cohort used the function (linear congruential random number generation) of Sun Work Shop C 5.0. The distributions of age, sex and utilization of the cohort were representative of the population. The data of residence and income were not available in the NHIRD datasets because the Bureau of National Health Insurance did not release such sensitive data.
Researchers who wish to access NHIRD datasets must sign a user agreement form indicating that they will obey related regulations and acknowledge the NHIRD in their publications. The study was approved by the Institutional Review Board of the Taipei Veterans General Hospital.
In the current study, we used only the ambulatory visit files of cohort datasets in 2002 (R{01..04}_CD2002.DAT). One record in the visit file might represent a consultation at the clinics or outpatient departments, a visit to emergency departments or preventive service, a prescription refill, or a use of ancillary services. There were a total of 2,572,065 visit records. One visit record contained up to three diagnostic codes in ICD-9-CM (International Classification of Diseases, Ninth Revision, Clinical Modification).
Besides, we used the registries for beneficiaries (R{01..04}_ID2002.DAT) and contracted healthcare facilities within the NHI (HOSB2002.DAT) to know the period of a person's eligibility for insurance and the accreditation category of healthcare facilities. We also used the registry for catastrophic illness patients (HV2002.DAT) to identify the patients with catastrophic illness. The updated registry for board-certified specialists (DOC2004.DAT) served to calculate the numbers of specialists per 100,000 people of the population.
Study design
Among all ambulatory visit records in 2002, we calculated only those visits with physician consultations of western medicine, dentistry, and traditional Chinese medicine, including visits to emergency departments. We used the consultation fee to differentiate between a visit with physician consultation and visits merely for radiology, laboratory examinations, physical rehabilitation, or other ancillary services. Prescription refills, home care by nurses, and preventive services without physician consultation would also be excluded from analysis.
We at first analyzed the utilization patterns of ambulatory care visits at different specialties and at different types of healthcare facilities. The statistics were displayed both in numbers of visits and in numbers of patients. Within the NHI, 24 specialties and 22 subspecialties were recognized. In general, the subspecialties existed only at hospitals. We grouped the family physicians and those practitioners without any specialist title into general practice; otherwise, a subspecialty was deemed different from its main specialty. A healthcare facility with physician services was contracted with the NHI in one of 4 categories: academic medical center, metropolitan hospital, local community hospital, and physician clinic. Besides, we also calculated the distribution of principal diagnoses at all visits by ICD-9-CM chapter.
For person-based analyses, we calculated the numbers of visits, consulted specialties, physicians and healthcare facilities by each person during 2002. Their distributions of frequencies would be displayed. We also calculated the age-sex distribution in each group of patients by annual visit count.
The patients with catastrophic illness and the visits due to catastrophic illness would be separately identified and integrated into the analysis.
To investigate how frequently a patient might change physicians and healthcare facilities, we calculated the numbers of consultations in which the patient had visited other healthcare facilities on the same day, in the past 3 days, and in the past 7 days, respectively. We also calculated the numbers of patients with such help-seeking behaviors during the year. Furthermore, we calculated the numbers of consultations in which the patient had visited the same specialty at other healthcare facilities within the same time frames. Finally, we would try to quantify the "one-stop shopping" phenomenon, in which a patient might visit several specialties at the same hospital in a day.
Data processing and statistical analysis
The open-source software Perl (version 5.8.6) [10] was used for computing. The regular statistics were displayed. The cohort originated from the people insured between 1995 and 2000. Because of death and emigration, not every person was still present in 2002. Our analysis was limited to the year 2002. In calculating the percentages of patients, the denominator was the number of persons who had been ever insured in 2002 according to the registry for beneficiaries.