The health benefits of fruit and vegetable (F&V) consumption are widely recognised. A high F&V intake has been associated with low risks of hypertension( Reference Wang, Manson and Gaziano 1 ), CHD( Reference Hung, Joshipura and Jiang 2 ), ischaemic heart disease( Reference Sharma, Vik and Kolonel 3 ) and mortality (all-cause, cancer and CVD)( Reference Nicklett, Semba and Xue 4 − Reference Zhang, Shu and Xiang 6 ) in community-dwelling elderly people. Conversely, high consumption of animal-derived food, such as red meat, particularly processed meat, is associated with a high risk of all-cause mortality( Reference Sinha, Cross and Graubard 7 − Reference Rohrmann, Overvad and Bueno-de-Mesquita 9 ).
Food security is a growing concern worldwide, and low-income households may face situations in which food or money is depleted. The global action plan for the prevention and control of non-communicable diseases (2013–2020) emphasises food availability, affordability and sustainability, contingent on optimising F&V consumption( 10 ). Food price is a major determinant of food choice( Reference Darmon and Drewnowski 11 ), and food expenditure correlates positively with a high intake of F&V and the associated dietary quality( Reference Lo, Chang and Lee 12 ).
The Taiwan National Health Insurance (NHI) is a compulsory programme launched in 1995; however, certain healthcare-related problems and challenges persist, particularly incremental ageing and non-communicable diseases( Reference Ho Chan 13 ). Personal healthcare expenditure for people aged 60 years or older is 2–3-fold that of the general population( 14 ). With ageing and disability attributable to non-communicable diseases, healthcare costs can further increase unless affordable preventive measures are taken.
Household or individual spending more on F&V rather than animal-derived foods has been associated with a reduced mortality risk for children under the age of 5 years in Indonesia( Reference Campbell, Thorne-Lyman and Sun 15 ) and for older adults in Taiwan( Reference Lo, Chang and Wahlqvist 16 ). This suggests that F&V expenditure has health outcome relevance across age groups. To our knowledge, the relationship between food group costs and those of the medical-care system is yet to be explored. Nutritional-economic approaches to address health outcomes are required, particularly for vulnerable groups such as children, reproductive-age women, older adults and low-income families. We have prospectively studied the effects of food group costs on medical service utilisation and costs in a nation-wide study in Taiwan of older adults.
Methods
Study participants
Free-living participants aged 65 years or older were recruited to a nation-wide representative Elderly Nutrition and Health Survey in Taiwan (NAHSIT) between 1999 and 2000. The study design has been published elsewhere( Reference Pan, Hung and Shaw 17 ). We interviewed 1911 participants through a home-based survey, which included a 24-h dietary recall. Trained interviewers used several tools such as food-piece models, abstract food models, measuring cups, spoons, electronic weighing and cue-cards to enquire about foods and dishes consumed in the past 24 h. In order to cross-validate the recall data, we also involved other members of the household( Reference Wu, Chang and Wei 18 ). To ensure the representativeness of the estimates for an individual diet, 126 participants with unreasonably high or low total daily energy intakes (<2092 kJ/d (<500 kcal/d) or >14 644 kJ/d (>3500 kcal/d) in women; <3347 kJ/d (<800 kcal/d) or 17 573 kJ/d (>4200 kcal/d) in men)( Reference Willett 19 ) and two participants on a liquid diet were excluded. The remaining participants were linked to the National Death Registry and NHI claim data up to 31 December 2006 (8-year follow-up), through their unique personal identification number. Of these, 133 participants had inaccurate death records, missing NHI records because of an incorrect personal identification number or a catastrophic illness before the interview; therefore, they were excluded. We did not wish to entertain the notion that high technology approaches and costs associated with terminal illness would be food sensitive. Participants who were older or died in the 1st year of follow-up might have had a prolonged illness or disease that manifested before death; these participants would have tended to use more medical services than others( Reference Yang, Norton and Stearns 20 ). Therefore, analysis was performed by further excluding participants who died in the 1st year of follow-up and those aged over and equal to 80 years (1445 participants left) (Fig. 1).
Flow chart for participant selection. NHI, National Health Insurance.
Ethics
The present study was approved by the ethics committees of both Academia Sinica and the National Health Research Institutes in Taiwan. All the participants provided their written informed consent.
Daily food group costs
Food retail prices between January 1999 and December 2000 were compiled using several databases from the Taiwanese Council of Agriculture – namely, the Agriculture and Food Agency, Poultry and Livestock Products Current Trade and the Fisheries Agency( 21 − 23 ). The national average monthly prices (New Taiwan dollars, NT$) for 843 foods (as purchased) were used to calculate the food costs at the interview dates. Prices were supplemented and adjusted according to the national supermarket chain price consistency or fluctuations throughout Taiwan in 2009. Supermarket food prices were discounted and deflated according to the consumer price index (CPI) for 1999–2000. Aggregate food costs (NT$/d) were calculated by multiplying the purchased weight (g) by the price of that particular food (NT$/g) and summing the cost of each food group or all foods an individual had consumed as per the 24-h dietary recall. We assumed that the participants ate at home, and no adjustments were made for eating out at restaurants, for personal celebrations or for festive occasions according to the Chinese or other calendar. We have published the food price calculations in detail( Reference Lo, Chang and Lee 12 ). In 1999–2000, US$ 1 equalled NT$31.
All the food items were classified into four groups: F&V comprising vegetables including fresh beans, fruits and fresh fruit juices; animal-derived foods comprising chicken, duck, pork, beef, goat, finfish, shellfish, crustaceans, eggs and dairy products; grain-based foods comprising rice and rice products, wheat and flour products, starchy roots and stems and dry beans; and miscellaneous foods and snacks except noodles and steamed buns. Thus, four food group costs were derived. Energy-adjusted food group costs were obtained using the residual method( Reference Willett 24 ) and assigned to tertiles (T).
National Health Insurance database
The NHI is a universal, compulsory health insurance programme financed by a means-related premium system launched in 1995, covering >99 % of the residents of Taiwan( Reference Cheng 25 , Reference Cheng 26 ). The National Health Insurance Research Database (NHIRD) provides claim data for reimbursement under the NHI. The NHIRD maintains complete beneficiary records comprising the clinical diagnoses (according to the International Classification of Diseases, Ninth Revision, Clinical Modification codes), catastrophic illnesses, medical costs and prescription claims associated with all outpatient and inpatient care( 27 ).
Medical service utilisation and costs
Medical service utilisation and costs were obtained from the NHI database. Both ambulatory care and hospital care utilisation and costs were considered and totalled. Ambulatory care included outpatient services, preventive care, dental services and emergency visits. Hospitalisation days were inpatient length of stay. Costs accounted for all resource inputs including those of medical and surgical treatments and contingent service fees. Medical data from the NHI claim data were collected longitudinally from the 1999−2000 interview date until the date of death or 31 December 2006. The average annualised utilisation and costs were obtained by dividing the total by the follow-up years for each participant. We assigned successive annual medical costs a yearly discount rate of 3 %, based on a yearly core CPI adjustment( Reference Lo, Wahlqvist and Chang 28 ). For reasons presented under study participants, medical services or incurred medical costs within 12 months of death were excluded.
Covariates
The potential covariates were demographic factors, socio-economic status (SES), multi-morbidity, physical function, physical activity, sleeping quality, chewing ability and nutritional factors( Reference Lo, Chang and Lee 12 , Reference Lo, Wahlqvist and Chang 28 − Reference Moriya, Tei and Harada 32 ). Responses were recorded using personal interviews, a personally administered questionnaire at baseline. Demographic and SES factors recorded were sex, age (years), regions, ethnicity (Fukienese, Hakka, mainlander and indigenous), education level (illiterate, primary and lower, secondary education and higher), living arrangement, self-rated financial status and household income (<15 000, 15 000−29 999, 30 000−49 999, ≥50 000 NT$/month). Health status was perceived as good, fair or poor. Physical function was documented on the basis of activities of daily living (ADL). A total of nine questions regarding self-care task difficulty were asked, which included eating, transfer between bed and chair, walking indoors/outdoors, dressing, bathing, toileting and urinary/bowel continence. Everyday tasks can reflect physical health and living independence( Reference Wiener, Hanley and Clark 33 ). One or more of these difficulties was considered a significant limitation in physical function( Reference Huang, Wahlqvist and Lee 34 ). Types and duration of sport and leisure time activities per week were obtained by the questionnaire, and daily physical activity as the number of metabolic equivalents (MET) was estimated. We classified physical activity into three levels (<1·5, 1·5–3, >3 MET/d)( Reference Huang, Wahlqvist and Lee 34 ). Sleeping quality was assessed using the following question: ‘How would you rate your overall sleeping quality in the past 3 months?’. The response was good, fair and poor( Reference Huang, Wahlqvist and Lee 35 ). Self-reported chewing ability was assessed using the following question: ‘Do you have difficulty chewing food?’. The response was either yes or no. For considering multi-morbidity, we used the Charlson co-morbidity index( Reference Charlson, Pompei and Ales 36 ) derived at baseline (1999 NHI claim data). Dietary quality was determined according to the 24-h recall and expressed as a dietary diversity score (DDS, range 0−6)( Reference Lee, Huang and Su 37 ).
Statistical analysis
The mean and median of medical service utilisation and cost data are reported because their distributions are skewed to the right. Categorical variables were compared using χ 2 tests. Generalised linear models (GLM) with a log link and γ distribution were used to identify the relationships between food group costs (tertiles, T) and medical costs without and with potential covariates while being simultaneously adjusted for total food, vegetable and fruit and animal-derived foods costs. For medical service utilisation, GLM with a log link and Poisson distribution were used( Reference Manning, Basu and Mullahy 38 ). The cost for respondents who spent no money on medical costs was substituted using NT$0·01. We used SAS, version 9.2 (SAS Institute Inc.) for all analyses. A two-tailed P<0·05 was considered to be statistically significant.
Results
Participants were followed-up for up to 8 years. Elderly participants whose F&V costs were in the T2 or T3 ranges had significantly lower annual average emergency (visits and costs), hospitalisation (days and costs) and total medical costs but had higher outpatient services and dental services (visits) compared with those in T1. In contrast, those whose animal-derived food cost was in the T2 or T3 range had significantly higher annual average emergency (visits), hospitalisation (days) and total medical costs compared with those in T1. The mean total medical costs for increasing tertiles of F&V costs were NT$67 500, NT$46 100 and NT$49 500; for increasing tertiles of animal-derived food costs, the costs were NT$50 000, NT$55 500 and NT$57 500. We focused on F&V and animal-derived food costs associated with medical service utilisation and costs on outpatient services, emergency, hospitalisation and total medical costs (Table 1).
Annual medical service utilisation and costs by tertile (T) of daily food group costs (n 1445) (Mean values and medians)
F&V, fruits and vegetables.
* P<0·05, ** P<0·01, *** P<0·001.
† Outcome was assessed by GLM with Poisson distribution, T1 as a reference group.
‡ NTD, with an exchange rate about NTD 31 to USD 1 in 2015.
§ Outcome was assessed by GLM with γ distribution, T1 as a reference group.
|| Food group costs were calculated as energy-adjusted food cost by residual method.
Older adults who spent less on F&V (T1) were indigenous, illiterate, of low financial status or of poor health status or they had poor ADL, unsatisfactory chewing ability or a low DDS (P<0·001) (Table 2).
Participant characteristics by tertile (T) of daily fruits and vegetables (F&V) costs (NT$/d) (n 1445) (Numbers and percentages; mean values with their standard deviations)
MET, metabolic equivalents.
* Food group costs were calculated as energy-adjusted food cost by residual method.
† Differences across T were tested using the χ 2 test.
‡ Differences across T were tested using ANOVA.
Table 3 lists the multivariable models and sensitivity analyses. No statistically significant differences were observed for emergency visits and costs. However, food group costs (F&V and animal-derived) and outpatient services (visits) and hospitalisation (days and costs) were significantly associated. In model 2, elderly participants in the T3 F&V food cost group had significantly fewer hospitalisation days (30 %) and total medical costs (19 %) but higher outpatient services (13 %) compared with those in T1 after controlling for all potential covariates. For animal-derived food costs, elderly participants in the T3 group had significantly more outpatient services (7 %), hospitalisation days (28 %), hospitalisation costs (83 %) and total medical costs (38 %) compared with those in the T1 group.
Generalised linear model (GLM) results for annual medical service utilisation and costs by tertile (T) of fruits and vegetables (F&V) and animal-derived food costs (NT$/d) (n 1445) (exp (β Coefficients) and 95 % confidence intervals)
exp, exponentiate; ADL, activities of daily living; CCI, Charlson co-morbidity index; DDS, dietary diversity score.
* P<0·05, ** P<0·01, *** P<0·001.
†P trend was significant.
‡ Food group costs were calculated as energy-adjusted food cost by residual method.
§ Outcome was assessed by GLM with Poisson distribution, T1 as a reference group.
║ Model 1: adjusted simultaneously for total food, fruits and vegetables and animal-derived food costs, age (years) and sex.
¶ Model 2: further adjusted for regions, ethnicity, personal education, living arrangement, self-rated financial status, household income (NT$/month), perceived health status, ADL, chewing ability, smoking, alcohol drinking, betel nut chewing, CCI, DDS and physical activity.
†† Outcome was assessed by GLM with γ distribution, T1 as a reference group.
Fig. 2 illustrates the annual adjusted mean of inpatient care (days and costs) according to the tertiles of F&V and animal-derived food costs after covariates were controlled for. Elderly participants with F&V food costs in T1 had the longest inpatient stays (11·5 d) and the highest costs (NT$88 400) as well as total medical costs (NT$116 000) compared with those in T2 and T3. However, participants with animal-derived food costs in T1 had the shortest mean inpatient stays (7·99 d) and the lowest costs (NT$42 200) as well as total medical costs (NT$81 500).
Annual adjusted mean hospitalisation (days and costs) and total medical costs by tertiles (T) of daily fruits and vegetable (F&V) and animal-derived food costs (n 1445). Models were adjusted simultaneously for total food, vegetable and fruit and animal-derived foods costs, age (years), sex, regions, ethnicity, personal education, living arrangement, self-rated financial status, household income (NT$/month), perceived health status, activities of daily living, smoking status, alcohol drinking, betel nut chewing, chewing ability, Charlson co-morbidity index, dietary diversity score and physical activity. (a) P
trends in F&V (
, days: P<0·001; 
, costs: P=0·062); in animal-derived foods (days: P<0·001; costs: P=0·032). (b) P
trends in F&V (costs: P=0·017); in animal-derived foods (costs: P=0·002).
We further examined, by combination, the simultaneous F&V and animal-derived food costs to explore the consumption pattern that could lower the medical-care burden. For inpatient care, older adults with F&V costs in T2 and T3 and those with animal-derived food costs in T1 had the shortest annual adjusted mean stays (5·78 d) and lowest hospitalisation costs (NT$38 600) as well as the lowest total medical costs (NT$75 800). Compared with the combined F&V (T1) and animal-derived food (T1) costs, the mean difference in annual total medical costs for F&V (T2 and T3) and animal-derived food (T1) was NT$45 200 (Fig. 3).
Annual adjusted mean hospitalisation (days and costs) and total medical costs by combining daily fruits and vegetable (F&V) and animal-derived food costs (NT$/d) (n 1445). Models were adjusted for total food costs, age (years), sex, regions, ethnicity, personal education, living arrangement, self-rated financial status, household income (NT$/month), perceived health status, activities of daily living, smoking status, alcohol drinking, betel nut chewing, chewing ability, Charlson co-morbidity index, dietary diversity score and physical activity. P
trends are all significant (a) , Days: P<0·001; , costs: P<0·001; (b) costs: P<0·001).
Discussion
We found that higher F&V and lower animal-derived food costs were associated with lower annual emergency (visits and costs), inpatient (days and costs) and total medical costs among community-dwelling elderly people. If older adults were to spend more on F&V but less on animal-derived food, medical costs might be reduced.
F&V consumption during older adulthood is associated with a reduced likelihood of critical chronic diseases( Reference Hung, Joshipura and Jiang 2 ). Most studies have reported that a higher F&V intake is protective against CVD( Reference Daviglus, Liu and Pirzada 39 , Reference Ellingsen, Hjerkinn and Seljeflot 40 ) and cancer( Reference Daviglus, Liu and Pirzada 39 , Reference Bradbury, Appleby and Key 41 ). In particular, large prospective cohort studies have reported that high F&V consumption is associated with a low risk of all-cause, cancer and CVD mortalities( Reference Zhang, Shu and Xiang 6 , Reference Bellavia, Larsson and Bottai 42 , Reference Oyebode, Gordon-Dseagu and Walker 43 ). F&V are rich in fibre, antioxidants (e.g. vitamins C and E, β carotene and other carotenoids, Se, Zn, flavonoids and polyphenols), minerals (e.g. K, Mg and Ca), folate and phytochemicals( Reference Uttara, Singh and Zamboni 44 − Reference Lako, Trenerry and Wahlqvist 46 ), and these may be some of the factors in combination that reduce the risk factors. Previous studies have argued that oxidative stress, through the initiation and progression of diseases such as CVD, hyperlipidaemia, hypertension, diabetes, cataracts, cancer, neurodegenerative diseases and ageing( Reference Uttara, Singh and Zamboni 44 , Reference Harasym and Oledzki 47 , Reference Knight 48 ), may be the most relevant mechanism involved; however, the pathways are likely to be multiple and complex. Moreover, the putative health effects of F&V consumption are highly evident in the overall dietary pattern and depend on the synergistic action and mutual interaction of pluripotential food components( Reference Charlson, Pompei and Ales 36 , Reference Harasym and Oledzki 47 , Reference Trichopoulou, Kouris-Blazos and Wahlqvist 49 ).
Although the health benefits of F&V consumption are well documented, the economic costs involved have not been explored in detail( Reference Lo, Chang and Wahlqvist 16 , Reference Lo, Wahlqvist and Chang 28 ). Food cost has the potential to understand dietary affordability, relevant to many vulnerable groups such as the elderly participants – for example, the dietary recommendations made by the World Health Organization( 10 ) encourage a daily consumption of five servings or a minimum of 400 g (excluding potatoes and other starchy tubers) of F&V to reduce the risk of non-communicable diseases. The implementation of this recommendation would be more realistic if its consequences for food cost in relation to medical costs were known.
Daviglus et al. ( Reference Daviglus, Liu and Pirzada 39 ) examined whether F&V intake among middle-aged men in the USA was related to Medicare expenses in their old age and revealed that F&V intakes were inversely associated with CVD- and cancer-related expenses as well as total expenses. The present study revealed that older adults who spent more on F&V had fewer mean annual hospitalisation days, costs and total medical costs. For emergency visits and costs, the trends were similar, although not significant. The mean consumption of F&V for T2 was 472 g and consistent with the WHO minimum recommendation. For both the USA (National Health and Nutrition Examination Survey; NHANES)( Reference Casagrande, Wang and Anderson 50 ) and Taiwan (NAHSIT for 1993−1996, 1999−2000 and 2005−2008) data, there were only slight changes in F&V intakes with adult age( Reference Wu, Pan and Yeh 51 ). There may be scope to improve F&V intakes in future by combining nutrition education with food price policies, which might contain support for F&V systems( Reference Darmon, Lacroix and Muller 52 ) and their accessibility.
In addition, the present study demonstrated that higher animal-derived food cost was associated with increased total medical costs. Animal-derived foods such as meat (red meat or processed meats) increase the risk of chronic diseases( Reference von Ruesten, Feller and Bergmann 53 ), CHD( Reference Micha, Wallace and Mozaffarian 54 ), type 2 diabetes( Reference Pan, Sun and Bernstein 55 , Reference Kim, Keogh and Clifton 56 ), colorectal cancer( Reference Norat, Bingham and Ferrari 57 ) and all-cause mortality( Reference Larsson and Orsini 58 ). Nevertheless, small amounts of meat have the potential to improve protein and micronutrient intakes, through interaction with plant nutrients, and with the bioavailability of Fe( Reference Rohrmann, Overvad and Bueno-de-Mesquita 9 ). Notably, poultry consumption has not been associated with all-cause mortality( Reference Rohrmann, Overvad and Bueno-de-Mesquita 9 ). Odegaard et al. ( Reference Odegaard, Koh and Yuan 59 ) reported that dietary patterns rich in pork-based dim sum and meat were not associated with cancer or all-cause mortality in Chinese Singaporean non-smokers. Moreover, several vegetarian diet-based studies have not reported an increased all-cause mortality among meat eaters or non-vegetarians compared with vegetarians( Reference Key, Appleby and Spencer 60 , Reference Kwok, Umar and Myint 61 ); however, a rather modest benefit associated with CVD outcomes was observed( Reference Toffanello, Inelmen and Minicuci 62 ). In our study, the highest F&V and the lowest animal-derived food cost group had the least hospitalisation days and lowest costs as well as total medical costs. This suggests that a food consumption pattern high in F&V but low in animal-derived food could benefit health and reduce medical costs. As expected (see Table 4), older people with higher F&V costs but low animal-derived food costs (F&V, T2+T3 and AD, T1) had lower intakes of fat, cholesterol and vitamin D, but higher intakes of dietary fibre and vitamins A and C, compared with those with other amounts of the combination of animal-derived and F&V food costs. The dietary spending combination of this group had one of the lowest total food costs; yet, its quality was not unfavourable, except for vitamin D (and probably vitamin B12, although not assessed). Moreover, the total protein energy percentage was 14·2 %, which is acceptable and equates to 1 g/kg body weight per d. There are problems with protein intakes (0·77 g/kg body weight per d) where both F&V and animal-derived expenditures are the lowest. Thus, a favourable diet need not be costly in order to save healthcare resources. Likewise, the amounts of food required to achieve these relatively favourable food and medical costs were modest.
Daily energy and protein intakes and nutrient densities by tertile (T) of daily fruits and vegetables (F&V) and animal-derived (AD) foods costs (NT$/d) (n 1445) (Numbers and percentages; mean values)
BW, body weight; DDS, dietary diversity score.
* Food costs were calculated as energy-adjusted food cost by residual method.
† Differences across T were tested by ANOVA.
‡ The percentage of total food costs.
Another possible explanation for these findings might be that those who purchase and eat less meat are more health conscious and have more favourable health outcomes – for example, older adults who spent more on F&V and less on animal-derived foods had lower rates of smoking, alcohol drinking and betel nut chewing, but higher physical activity and better sleeping quality. Despite these covariates, the findings are unchanged. It also has to be noted that, in our study, when both F&V and animal-derived food intakes are low, the medical costs are high, consistent with our previous findings that show a benefit of dietary diversity for medical costs( Reference Lo, Wahlqvist and Chang 28 ). Our findings are not associated with food avoidance, but with moderation of animal-derived food intake.
Our findings are in agreement with the WHO recommendations for F&V consumption and draw attention to operationalising the recommendations regarding food category affordability and medical costs. In other words, public health nutrition policies require a nutritional-economic dimension.
Limitations
First, although our study was prospective, food group expenditure assessment was performed only at commencement. Thus, we have no information about changes in the eating behaviour or long-term dietary habits of the participants. However, community-dwelling elderly people have relatively stable diets( Reference Lee, Huang and Su 37 , Reference Toffanello, Inelmen and Minicuci 62 ), and the reproducibility (intra-class correlation coefficient) for various nutrients of the three repeated 24-h dietary recalls for a subgroup population of fifty averaged 0·46( Reference Huang, Wahlqvist and Kao 63 ). Second, food prices were derived from the national average monthly prices for Taiwan, and regional price variations were not considered( Reference Clauson and McCray 64 ). Third, food prices were based on raw materials and did not include the costs of preparing food or eating out. This could have resulted in overestimation or underestimation of the actual food costs. Nevertheless, several studies have successfully used a similar food-pricing approach( Reference Lo, Chang and Lee 12 , Reference Lo, Chang and Wahlqvist 16 , Reference Lo, Wahlqvist and Chang 28 ). Finally, although the NHI covers >99 % of the population of Taiwan, out-of-pocket (OOP) payments are required for some medical-care services, pharmaceuticals and devices not covered by the NHI programme. In Taiwan, household income is positively associated with the use of healthcare and OOP payments( Reference Chi, Lee and Tsai 30 ). OOP healthcare expenditures increase per household at all income levels; however, they constitute a substantial economic burden for low-income households( Reference Sanmartin, Hennessy and Lu 31 ). Therefore, the medical costs of high-income households may have been underestimated. Moreover, social disparities remain prevalent in healthcare accessibility and affordability, causing the medical service use and costs of low-income households to be underestimated. To address some of these limitations, we adjusted for SES and region.
Conclusions
F&V cost is inversely associated with hospitalisation (days and costs) and total medical costs. Animal-derived food cost is positively associated with hospitalisation and medical costs among Taiwanese free-living older adults. A high F&V and low animal-derived food spending pattern could lower the economic burden of the medical-care system.
Acknowledgements
The data sets used were obtained from the Elderly Nutrition and Health Survey in Taiwan, 1999−2000, conducted by the Center for Survey Research, Academia Sinica, and directed by Wen-Harn Pan and Su-Hao Tu. This study was partly based on data from the National Health Insurance Research Database provided by the Bureau of National Insurance, Department of Health, and managed by the National Health Research Institutes. The interpretation and conclusions of this study do not represent those of the Bureau of National Health Insurance, Department of Health or National Health Research Institutes.
Financial support for this project and for the research fellowship of Y. -T. C. L. was provided by the Ministry of Science and Technology (MOST103-2320-B-016-015-MY2, MOST103-2811-B-016-003, MOST104-2811-B-016-007).
All the authors were involved in the study design and collectively wrote the paper; Y. -T. C. L. analysed the data; Y.-C. H. contributed to the interpretation of the results; M. L. W. and M.-S. L. coordinated the research; and M.-S. L. had the primary responsibility for the final content.
None of the authors has any conflicts of interest to declare.
