To understand methodological concerns related to dietary intake collection in transitional societies.

Three days of household weighing and measurement of all food used and repeated 24-hour recalls.

Eight Chinese provinces.

Five thousand nine hundred and fifty-two and 5152 adults aged 20–45 years in 1989 and 1997, respectively.

Great variance exists in the types and quantity of animal products used in common recipes. For example, the proportion of pork from lean cuts in ‘stir-fried fresh pepper and pork’ varies between 14 and 24% in urban and rural areas and the total pork content for 100 g (dish) varies by 15 to 19 g between rural and urban areas in each of eight provinces. Another challenge relates to the variation in the edible vegetable oil content added during food preparation. Reliance on standard recipes for each fried dish would miss the variations in oil use over time, space and socio-economic status.

Dietary change is rapid in transitional countries. Reliance on recipes standardised for animal food and edible oil contents will lead to very large systematic errors in the measurement of energy, fat and protein intakes.

To examine statistical models that account for correlation between random errors of different dietary assessment methods, in dietary validation studies.

In nutritional epidemiology, sub-studies on the accuracy of the dietary questionnaire measurements are used to correct for biases in relative risk estimates induced by dietary assessment errors. Generally, such validation studies are based on the comparison of questionnaire measurements (Q) with food consumption records or 24-hour diet recalls (R). In recent years, the statistical analysis of such studies has been formalised more in terms of statistical models. This made the need of crucial model assumptions more explicit. One key assumption is that random errors must be uncorrelated between measurements Q and R, as well as between replicate measurements R1 and R2 within the same individual. These assumptions may not hold in practice, however. Therefore, more complex statistical models have been proposed to validate measurements Q by simultaneous comparisons with measurements R plus a biomarker M, accounting for correlations between the random errors of Q and R.

The more complex models accounting for random error correlations may work only for validation studies that include markers of diet based on physiological knowledge about the quantitative recovery, e.g. in urine, of specific elements such as nitrogen or potassium, or stable isotopes administered to the study subjects (e.g. the doubly labelled water method for assessment of energy expenditure). This type of marker, however, eliminates the problem of correlation of random errors between Q and R by simply taking the place of R, thus rendering complex statistical models unnecessary.

To provide an account of the state of diet and chronic disease research designs and methods; to discuss the role and potential of aggregate and analytical observational studies and randomised controlled intervention trials; and to propose strategies for strengthening each type of study, with particular emphasis on the use of nutrient biomarkers in cohort study settings.

Observations from diet and disease studies conducted over the past 25 years are used to identify the strengths and weaknesses of various study designs that have been used to associate nutrient consumption with chronic disease risk. It is argued that a varied research programme, employing multiple study designs, is needed in response to the widely different biases and constraints that attend aggregate and analytical epidemiological studies and controlled intervention trials. Study design modifications are considered that may be able to enhance the reliability of aggregate and analytical nutritional epidemiological studies. Specifically, the potential of nutrient biomarker measurements that provide an objective assessment of nutrient consumption to enhance analytical study reliability is emphasised. A statistical model for combining nutrient biomarker data with self-report nutrient consumption estimates is described, and related ongoing work on odds ratio parameter estimation is outlined briefly. Finally, a recently completed nutritional biomarker study among 102 postmenopausal women in Seattle is mentioned. The statistical model will be applied to biomarker data on energy expenditure, urinary nitrogen, selected blood fatty acid measurements and various blood micronutrient concentrations, and food frequency self-report data, to identify study subject characteristics, such as body mass, age or socio-economic status, that may be associated with the measurement properties of food frequency nutrient consumption estimates. This information will be crucial for the design of a potential larger nutrient biomarker study within the cohort study component of the Women's Health Initiative.

The methodology under study is expected to be pertinent to a wide variety of diet and chronic disease association studies in the general population. Ongoing work focuses on statistical methods developed using computer simulations motivated by studies of dietary fat in relation to breast and colon cancer among post-menopausal women, and ongoing pilot studies to be described in detail elsewhere, involving post-menopausal women living in the Seattle area.

A varied research programme appears to be needed to make progress in the challenging diet and chronic disease research area. Such progress may include aggregate studies of diet and chronic disease that include sample surveys in diverse population groups world-wide, analytical epidemiological studies that use nutrient biomarker data to calibrate self-report nutrient consumption estimates, and randomised controlled intervention trials that arise from an enhanced infrastructure for intervention development. New innovative designs, models and methodologies are needed for each such research setting.

To demonstrate that dietary datasets from the Continuing Survey of Food Intakes by Individuals, a US population survey, allow comparisons with national data and provide food composition datasets that can be used to generate similar dietary data.

Two studies are described: the Lower Mississippi Delta Nutrition Intervention Research Initiative (Delta NIRI), which used a 24-hour recall, and a Department of Defense Military Nutrition Research Task, which used 3-day dietary records. Both studies used the same food composition tables.

Rural Lower Mississippi Delta and an Army post.

Four hundred and nine residents (adults and children) from the rural Delta region of Arkansas, Louisiana and Mississippi, and 74 career soldiers from the Sergeants Major Academy, Fort Bliss, Texas.

The Delta NIRI study found that fruit and vegetable consumption for these rural residents was lower than that found nationally. Additionally, the quality of vegetable servings is of concern since a large percentage came from french fries and potato chips. In the Sergeants Major Academy study, the national survey food composition tables allowed for easy analysis of intake data and comparisons with dietary recommendations.

Strategies similar to those used for the Delta NIRI and Military Nutrition Research Task can be used widely, allowing comparisons of ‘defined populations’ with nationally distributed data. Additionally, measurement of dietary change is more efficient when the same protocol is used subsequently to collect more data, a method similar to that used by the US Department of Agriculture to describe food consumption patterns from one survey to another.

To share experience on the development of national and regional food composition tables (FCTs) and the issues and problems involved.

Food composition data analysed at national and regional levels were compiled. Systematic development of FCTs detailing specific criteria and characteristics following the INFOODS and ASEANFOODS guidelines was conducted.

The national FCT was developed using the Thai FCT as a study model. For the regional level, the Association of South East Asian Nations (ASEAN) FCT was developed.

About 3000 analysed datasets of about 1000 food items on proximate composition, minerals, vitamins and other available components from various sources in Thailand and about 4500 analysed datasets of 1700 food items on the same components from various countries in ASEAN were compiled. The national and regional FCTs were then developed systematically.

The developed Thai FCT contains 24 nutrients of 1055 food items and the ASEAN FCT contains 21 nutrients of about 1740 food items. Issues and problems regarding the development and use of national and regional food composition data are listed.

Food composition databases have become increasingly important in the international health arena. Systematic compilation of the data is necessary and the use of INFOODS tag names is strongly recommended. This is to indicate the methods of nutrient analysis and calculation, and to facilitate international and regional data interchange. Users are also recommended to read the Explanatory Notes and Information to the Users. Collaboration among data generators, users and compilers at national and regional levels, as well as support from international organisations, is essential.