- Email: [email protected]
Quality control of nutrient data entry for a long-term, multi-centre dietary intervention trial
Journal of Food Composition and Analysis 22S (2009) S88–S92
Contents lists available at ScienceDirect
Journal of Food Composition and Analysis journal homepage: www.elsevier.com/locate/jfca
Short Communication
Quality control of nutrient data entry for a long-term, multi-centre dietary intervention trial Cary Greenberg *, Lorraine Gougeon, Barbara Selley, Valentina Kriukov, Qing Li, Norman F. Boyd, Lisa J. Martin University Health Network, Epidemiology and Statistics, 610 University Ave, 10th Floor, Room 10-407, Toronto, Ontario, Canada
A R T I C L E I N F O
A B S T R A C T
Article history: Received 24 September 2008 Received in revised form 21 January 2009 Accepted 24 January 2009
The Diet and Breast Cancer Prevention Study is a multi-centre, randomized intervention trial to determine if a low-fat, high-carbohydrate diet would reduce the incidence of breast cancer in women at increased risk for the disease. The study began in 1988 and was completed in December 2005. We recruited 4690 women from 5 cities across Canada and asked them to provide 3-day food records at each clinic visit. We analyzed 82,000 food records using Nutrition Data System from the Nutrition Coordinating Centre, University of Minnesota. Our quality control program needed to maximize accuracy and consistency despite a large staff in various locations, changes to the NDS that excluded Canadian foods and an astounding number of new food products. We screened prospective staff to assess their food knowledge and provided extensive training and continuing education. We compiled a supplementary database of new product information and data entry rules. Records with energy and fat intakes beyond usual thresholds were checked. Inter-dietitian variation was assessed by re-entering a sample of food records and achieved a correlation of greater than 0.9 for all macronutrients. This is the only study of diet and breast cancer prevention to collect food records and blood samples over a prolonged period. High quality nutrient data allows us to examine diet and breast cancer risk and phenotype. ß 2009 Elsevier Inc. All rights reserved.
Keywords: Diet and breast cancer prevention Nutrient data Quality control Food records Canada Dietary assessment Food data management Food data compilation Food composition
1. Introduction The Diet and Breast Cancer Prevention Trial was a long-term, multi-centre, randomized controlled study to determine if a lowfat, high-carbohydrate diet would reduce the incidence of breast cancer in healthy women, at increased risk of the disease due to extensive breast density (Boyd et al., 1995). The study began in 1988 and we recruited 4690 women from 5 study centres in Ontario and British Columbia. Subjects were randomly assigned to either the intervention or control group and follow up of all study participants continued until December 2005. Dietary data were collected using 3-day food records at baseline and at intervals throughout the trial. We have analyzed 82,000 food record days to date. Food records were used to assess the individual intake of the participants because they provided a more detailed account of eating habits than other dietary assessment methods. The specific information collected from food records was particularly useful when counselling participants in the intervention group. Food
records allowed us to detect deviations from the prescribed diet and provided insight into food preferences that enabled dietitians to suggest practical and useful strategies to improve adherence. Although there are limitations and burdens associated with them, it has been shown that estimates of nutrient intakes from food records have a higher validity than food frequency questionnaires (Bingham et al., 2003; Freedman et al., 2006). Quality control was considered essential to maximize the accuracy and consistency of nutrient data collection and analysis, especially in view of the large amount of nutrient data and the unique contribution our research could provide to the field of breast cancer prevention. In this paper we describe our quality control program for nutrient data entry that was based on hiring and training skilled dietitians and nutritionists, providing information tools to standardize the data entry process, and reviewing and evaluating our process. 2. Methods and results 2.1. Overview of trial
* Corresponding author. Tel.: +1 416 946 2059; fax: +1 416 946 2024. E-mail address: [email protected] (C. Greenberg). 0889-1575/$ – see front matter ß 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.jfca.2009.01.011
Women with greater than 50% breast density as seen on their mammogram were recruited from mammography units in
C. Greenberg et al. / Journal of Food Composition and Analysis 22S (2009) S88–S92
Toronto, Hamilton, London, Windsor and Vancouver. Potential participants were contacted by mail and invited to attend two screening visits to learn more about the study and how to keep food records. The instruction about record keeping began at the first screening visit when the dietitian completed a 24-h recall with the potential participant to demonstrate the type of information required on a food record. At the end of the first interview the participant was given a food record booklet and asked to complete 3 days of food records before the second screening visit, which took place about 2 weeks later. The food record booklet included instructions and guidelines about documenting product information, preparation techniques, recipes and measurements according to the different types of foods or beverages consumed. Each food record sheet in the booklet also included a ruler printed along the side of the page to facilitate the documentation of serving sizes when measuring with a scale or measuring cups or spoons was not feasible. Eligible subjects who provided food records of adequate quality and gave consent were randomly assigned to one of the two study groups and then given a scale to be used for food records. Women in the control group were given a copy of Canada’s Food Guide and were not asked to make any changes to their fat intake. They were seen every 3 months in the first year, every 6 months in the second year and then annually. Women in the intervention group were given individualized instruction to achieve and maintain a low-fat, high-carbohydrate diet that was based on an isocaloric replacement of fat with carbohydrate with a target energy distribution of 15% fat, 65% carbohydrate and 20% protein. The intervention was designed to reduce fat and increase carbohydrate while emphasizing grains, vegetables and fruit, limiting animal protein foods and maintaining body weight. Subjects in the intervention group were seen monthly for the first year, every 3 months in the second year, and every 6 months thereafter. All subjects were followed for 7–17 years and for an average of 10 years for the development of breast cancer or other outcomes. At baseline and every visit thereafter, subjects in both groups were asked to provide food records on three non-consecutive days, selected by the dietitian. All food records were reviewed at the clinic visit by a dietitian to ensure adequate detail and clarity; and portion sizes were verified with food models when necessary. Dietitians analyzed the records without knowledge of the identity and study group of the subjects. They used the Nutrition Data System (NDS Versions 2.1–2.91) developed by the Nutrition Coordinating Centre at the University of Minnesota. 2.2. Quality control methods 2.2.1. Screening and training of nutrient data entry staff Over the 17-year duration of the trial, 46 dietitians and nutritionists performed nutrient data entry in five different study sites in Ontario and British Columbia. To preserve the integrity of the nutrient data it was very important to hire and train highly qualified staff. Dietitians and nutritionists were required to have extensive knowledge of cooking methods and domestic and imported foods to efficiently and accurately analyze food records from thousands of participants with diverse preferences and eating habits. Prospective candidates were screened using a short knowledge test that asked about measurements used in cooking, commercial and ethnic foods and nutrition composition (see Fig. 1 for the test questions). The nutrient data manager (LG) was responsible for the rigorous training and certification program to ensure that staff developed proficiency in the use of NDS software, followed uniform procedures and demonstrated good judgment for data entry of foods that did not have a direct match in the database. Each dietitian and nutritionist was closely supervised for about 3
S89
Fig. 1. Food knowledge test.
months or until 100 records were entered. During this time the nutrient data manager checked their work regularly and gave feedback for all data entry errors. After this comprehensive training program, most staff were performing at a high level of competency with very few food selection or typographical errors. In cases when the nutrient data manager felt additional supervision was required, their training continued until their work was considered acceptable. In addition, the nutrient data manager provided continuing education to all nutrient data entry staff at annual dietitian meetings to deal with challenges to data entry posed by changing trends in eating habits. 2.2.2. Standardizing nutrient data entry Although NDS has a very extensive database including some Canadian products, it could not include every food consumed by our study participants. Therefore, from the outset a supplementary database was compiled of these additional food items called the ‘‘Missing Food Book’’ (MFB). The MFB standardized the entry of new items and maximized accuracy, efficiency and consistency. Over time, tremendous changes occurred in the marketplace and reliance on restaurant meals was increasing. The number and variety of convenience and ethnic foods and foods modified in fat and fibre were increasing and popular and needed to be accommodated in the data entry system. We also needed to deal with changes to NDS that occurred in 1998 when a new version removed Canadian foods. We decided to continue with Version 2.91 and incorporate information about new products into the MFB. Dietitians and nutritionists recorded information about new products on a ‘‘Missing Food Form’’. This standardized form helped to ensure that the data necessary to understand and guide the entry for new products was collected. It included: product name, source of the nutrition information, i.e. whether it came from the label, website or telephone contact with company, package size and format, preparation instructions for mixes, main ingredients, nutritional composition and NDS entry required to closely match the nutritional content. A sample form is shown in Fig. 2. The format of the MFB also progressed so that it could be updated relatively quickly and easily. The final version was developed using two Microsoft Office programs, Access and Excel. Access was used to assemble and organize the product database
S90
C. Greenberg et al. / Journal of Food Composition and Analysis 22S (2009) S88–S92
Fig. 2. Missing food form.
and data entry rules. The specific product name, including flavour/variety if applicable, was listed along with the NDS entry for a particular serving size. The database also included common package and unit sizes for each item to guide entry when the quantity consumed differed from the reference serving size.
Data entry rules included guidelines for situations when descriptions and/or portions were not specified on the food record, raw to cooked conversions of vegetables and fruits, average sizes and weights of pasta and fresh produce, entering sushi and ribs and calculating volumes of soup. The database
C. Greenberg et al. / Journal of Food Composition and Analysis 22S (2009) S88–S92
S91
Table 1 Mean nutrient values for initial and repeat entry. Overall
Mean initial entry (N = 79)
Mean repeat entry (N = 79)
P-value of differencea
Energy (kJ) Total fat (g) Total protein (g) Total carbohydrate (g)
6774 (1581) 49.0 (20.8) 66.5 (13.7) 227.7 (63.9)
6769 (1530) 48.8 (20.7) 65.0 (13.4) 230.4 (61.0)
0.91 0.73 0.01 0.10
Group = control
Mean initial entry (N = 38)
Mean repeat entry (N = 38)
P-value of differencea
Energy (kJ) Total fat (g) Total protein (g) Total carbohydrate (g)
7288 (1556) 63.5 (18.7) 66.8 (13.2) 223.1 (52.8)
7306 (1566) 63.2 (19.1) 66.0 (13.5) 226.6 (51.9)
0.78 0.79 0.36 0.07
Group = intervention
Mean initial entry (N = 41)
Mean repeat entry (N = 41)
P-value of differencea
Energy (kJ) Total fat (g) Total protein (g) Total carbohydrate (g)
6299 (1468) 35.6 (11.8) 66.3 (14.4) 232.1 (73.0)
6271 (1329) 35.4 (10.8) 64.1 (13.3) 233.9 (68.8)
0.68 0.84 0.009 0.48
a
P-value from one sample t-test.
was exported to Excel for distribution to staff on a quarterly basis. 2.2.3. Review and evaluation of nutrient data entry A review process was established from the outset of the trial to detect nutrient data entry errors. Time and cost constraints made it impossible to check every record that was analyzed so criteria were established to identify records where errors might be more likely to occur. The nutrient data manager checked records when daily energy intake was less than or equal to 800 cal, greater than
or equal to 2800 cal or total fat intake was greater than or equal to 110 g. These limits correspond to about three standard deviations from the mean for energy and fat which was consistent with the criteria used in the Multiple Risk Factor Intervention Trial (Tillotson et al., 1986). The impact on workload for the nutrient data manager of the limits established for ‘‘out of range’’ records was that 10–12% food records needed to be checked and 35% of the records reviewed were changed. To evaluate consistency in nutrient analysis between dietitians two studies were conducted where food records were re-entered
Fig. 3. Repeated entry with initial entry.
S92
C. Greenberg et al. / Journal of Food Composition and Analysis 22S (2009) S88–S92
by a different dietitian. In 1994, 140 randomly selected food records, which represented 10% of the total food records analyzed by that time, were re-entered. There were no differences in the estimates of macronutrients expressed in grams or percent of energy between the first and second analysis when expressed as percent difference. Correlations between all macronutrients were high (mean correlation coefficient = 0.93) (Martin et al., 1994) In 2002 a second re-entry study was conducted to determine the reliability of data entry using different versions of NDS. As mentioned earlier, after 1998 updated versions of NDS were not used so it was important to determine the reliability of entry using early (2.1–2.7) and a later Version (2.91) of NDS. A total of 79 randomly selected 3-day food record sets (38 sets from the control group, 41 sets from the intervention group) that were previously entered in the earlier versions were re-entered in Version 2.91. The subject’s name, study group and study centre were concealed on the food record and distributed to all available dietitians. The results of the second double entry experiment are presented in Table 1 and Fig. 3. There were no significant differences in the means between the initial and repeat entry for energy and total fat and carbohydrate, overall and in the intervention and control groups. There was a very small but statistically significant difference in the grams of total protein between the initial and repeat entry, overall and in the intervention group. Correlations between initial and repeat entries were all greater than 0.9 for the overall, and were similar in the intervention and control groups. This shows strong agreement in nutrient data when food records are entered by different dietitians and different versions of the NDS database and provides convincing evidence that our nutrient analysis system is highly reliable. 3. Discussion and conclusions The nutrient data for the Diet and Breast Cancer Prevention Study are derived from 3-day food records collected from a large
number of subjects over many years. Limitations for all nutritional assessment techniques are well recognized but incorporating quality control procedures for nutrient data entry into study procedures minimized some of the measurement error associated with food records (Prentice, 1996). As a result, high quality nutrient data are available for examination of the relationship between diet and breast cancer risk and phenotype. Designing an effective quality control system for a long-term, multi-centre trial with a large staff needs to be multi-faceted and continual to cope with the magnitude of the task. Selecting highly qualified professionals, providing them with comprehensive training and continuing education, implementing on-going monitoring activities and developing strategies to respond to problems in a timely manner resulted in extremely high quality performance. Furthermore, these processes could easily be applied to single-centre or smaller trials that have similar concerns regarding achieving and maintaining quality data. References Bingham, S.A., Luben, R., Welch, A., Wareham, N., Khaw, K.-T., Day, N., 2003. Are imprecise methods obscuring a relation between fat and breast cancer? Lancet 362, 212–214. Boyd, N.F., Fishell, E., Jong, R., MacDonald, J.C., Sparrow, R.K., Simor, I.S., Kriukov, V., Lockwood, G., Tritchler, D., 1995. Mammographic densities as a criterion for entry to a clinical trial of breast cancer prevention. Br. J. Cancer 72, 476– 479. Freedman, L.S., Potischman, N., Kipnis, V., Midthune, D., Schatzkin, A., Thompson, F.E., Troiano, R.P., Prentice, R., Patterson, R., Carroll, R., Subar, A.F., 2006. A comparison of two dietary instruments for evaluating the fat–breast cancer relationship. Int. J. Epidemiol. 35, 1011–1021. Martin, L.J., Cousins, M.L., Gougeon, L.L., Kruikov, V., Lockwood, G.A., Boyd, N.F., 1994. Quality control of nutrient analysis of food records in a low fat intervention trial. Am. J. Clin. Nutr. 59, 287S. Prentice, R.L., 1996. Measurement error and results from analytic epidemiology: dietary fat and breast cancer. J. Natl. Cancer Inst. 88, 1738–1747. Tillotson, J.L., Gorder, D.D., Duchene, A.G., Grambsch, P.V., Wenz, J., 1986. Quality control in the multiple risk factor intervention trial nutrition modality. Control Clin. Trials 7, 66S–90S.
Contents lists available at ScienceDirect
Journal of Food Composition and Analysis journal homepage: www.elsevier.com/locate/jfca
Short Communication
Quality control of nutrient data entry for a long-term, multi-centre dietary intervention trial Cary Greenberg *, Lorraine Gougeon, Barbara Selley, Valentina Kriukov, Qing Li, Norman F. Boyd, Lisa J. Martin University Health Network, Epidemiology and Statistics, 610 University Ave, 10th Floor, Room 10-407, Toronto, Ontario, Canada
A R T I C L E I N F O
A B S T R A C T
Article history: Received 24 September 2008 Received in revised form 21 January 2009 Accepted 24 January 2009
The Diet and Breast Cancer Prevention Study is a multi-centre, randomized intervention trial to determine if a low-fat, high-carbohydrate diet would reduce the incidence of breast cancer in women at increased risk for the disease. The study began in 1988 and was completed in December 2005. We recruited 4690 women from 5 cities across Canada and asked them to provide 3-day food records at each clinic visit. We analyzed 82,000 food records using Nutrition Data System from the Nutrition Coordinating Centre, University of Minnesota. Our quality control program needed to maximize accuracy and consistency despite a large staff in various locations, changes to the NDS that excluded Canadian foods and an astounding number of new food products. We screened prospective staff to assess their food knowledge and provided extensive training and continuing education. We compiled a supplementary database of new product information and data entry rules. Records with energy and fat intakes beyond usual thresholds were checked. Inter-dietitian variation was assessed by re-entering a sample of food records and achieved a correlation of greater than 0.9 for all macronutrients. This is the only study of diet and breast cancer prevention to collect food records and blood samples over a prolonged period. High quality nutrient data allows us to examine diet and breast cancer risk and phenotype. ß 2009 Elsevier Inc. All rights reserved.
Keywords: Diet and breast cancer prevention Nutrient data Quality control Food records Canada Dietary assessment Food data management Food data compilation Food composition
1. Introduction The Diet and Breast Cancer Prevention Trial was a long-term, multi-centre, randomized controlled study to determine if a lowfat, high-carbohydrate diet would reduce the incidence of breast cancer in healthy women, at increased risk of the disease due to extensive breast density (Boyd et al., 1995). The study began in 1988 and we recruited 4690 women from 5 study centres in Ontario and British Columbia. Subjects were randomly assigned to either the intervention or control group and follow up of all study participants continued until December 2005. Dietary data were collected using 3-day food records at baseline and at intervals throughout the trial. We have analyzed 82,000 food record days to date. Food records were used to assess the individual intake of the participants because they provided a more detailed account of eating habits than other dietary assessment methods. The specific information collected from food records was particularly useful when counselling participants in the intervention group. Food
records allowed us to detect deviations from the prescribed diet and provided insight into food preferences that enabled dietitians to suggest practical and useful strategies to improve adherence. Although there are limitations and burdens associated with them, it has been shown that estimates of nutrient intakes from food records have a higher validity than food frequency questionnaires (Bingham et al., 2003; Freedman et al., 2006). Quality control was considered essential to maximize the accuracy and consistency of nutrient data collection and analysis, especially in view of the large amount of nutrient data and the unique contribution our research could provide to the field of breast cancer prevention. In this paper we describe our quality control program for nutrient data entry that was based on hiring and training skilled dietitians and nutritionists, providing information tools to standardize the data entry process, and reviewing and evaluating our process. 2. Methods and results 2.1. Overview of trial
* Corresponding author. Tel.: +1 416 946 2059; fax: +1 416 946 2024. E-mail address: [email protected] (C. Greenberg). 0889-1575/$ – see front matter ß 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.jfca.2009.01.011
Women with greater than 50% breast density as seen on their mammogram were recruited from mammography units in
C. Greenberg et al. / Journal of Food Composition and Analysis 22S (2009) S88–S92
Toronto, Hamilton, London, Windsor and Vancouver. Potential participants were contacted by mail and invited to attend two screening visits to learn more about the study and how to keep food records. The instruction about record keeping began at the first screening visit when the dietitian completed a 24-h recall with the potential participant to demonstrate the type of information required on a food record. At the end of the first interview the participant was given a food record booklet and asked to complete 3 days of food records before the second screening visit, which took place about 2 weeks later. The food record booklet included instructions and guidelines about documenting product information, preparation techniques, recipes and measurements according to the different types of foods or beverages consumed. Each food record sheet in the booklet also included a ruler printed along the side of the page to facilitate the documentation of serving sizes when measuring with a scale or measuring cups or spoons was not feasible. Eligible subjects who provided food records of adequate quality and gave consent were randomly assigned to one of the two study groups and then given a scale to be used for food records. Women in the control group were given a copy of Canada’s Food Guide and were not asked to make any changes to their fat intake. They were seen every 3 months in the first year, every 6 months in the second year and then annually. Women in the intervention group were given individualized instruction to achieve and maintain a low-fat, high-carbohydrate diet that was based on an isocaloric replacement of fat with carbohydrate with a target energy distribution of 15% fat, 65% carbohydrate and 20% protein. The intervention was designed to reduce fat and increase carbohydrate while emphasizing grains, vegetables and fruit, limiting animal protein foods and maintaining body weight. Subjects in the intervention group were seen monthly for the first year, every 3 months in the second year, and every 6 months thereafter. All subjects were followed for 7–17 years and for an average of 10 years for the development of breast cancer or other outcomes. At baseline and every visit thereafter, subjects in both groups were asked to provide food records on three non-consecutive days, selected by the dietitian. All food records were reviewed at the clinic visit by a dietitian to ensure adequate detail and clarity; and portion sizes were verified with food models when necessary. Dietitians analyzed the records without knowledge of the identity and study group of the subjects. They used the Nutrition Data System (NDS Versions 2.1–2.91) developed by the Nutrition Coordinating Centre at the University of Minnesota. 2.2. Quality control methods 2.2.1. Screening and training of nutrient data entry staff Over the 17-year duration of the trial, 46 dietitians and nutritionists performed nutrient data entry in five different study sites in Ontario and British Columbia. To preserve the integrity of the nutrient data it was very important to hire and train highly qualified staff. Dietitians and nutritionists were required to have extensive knowledge of cooking methods and domestic and imported foods to efficiently and accurately analyze food records from thousands of participants with diverse preferences and eating habits. Prospective candidates were screened using a short knowledge test that asked about measurements used in cooking, commercial and ethnic foods and nutrition composition (see Fig. 1 for the test questions). The nutrient data manager (LG) was responsible for the rigorous training and certification program to ensure that staff developed proficiency in the use of NDS software, followed uniform procedures and demonstrated good judgment for data entry of foods that did not have a direct match in the database. Each dietitian and nutritionist was closely supervised for about 3
S89
Fig. 1. Food knowledge test.
months or until 100 records were entered. During this time the nutrient data manager checked their work regularly and gave feedback for all data entry errors. After this comprehensive training program, most staff were performing at a high level of competency with very few food selection or typographical errors. In cases when the nutrient data manager felt additional supervision was required, their training continued until their work was considered acceptable. In addition, the nutrient data manager provided continuing education to all nutrient data entry staff at annual dietitian meetings to deal with challenges to data entry posed by changing trends in eating habits. 2.2.2. Standardizing nutrient data entry Although NDS has a very extensive database including some Canadian products, it could not include every food consumed by our study participants. Therefore, from the outset a supplementary database was compiled of these additional food items called the ‘‘Missing Food Book’’ (MFB). The MFB standardized the entry of new items and maximized accuracy, efficiency and consistency. Over time, tremendous changes occurred in the marketplace and reliance on restaurant meals was increasing. The number and variety of convenience and ethnic foods and foods modified in fat and fibre were increasing and popular and needed to be accommodated in the data entry system. We also needed to deal with changes to NDS that occurred in 1998 when a new version removed Canadian foods. We decided to continue with Version 2.91 and incorporate information about new products into the MFB. Dietitians and nutritionists recorded information about new products on a ‘‘Missing Food Form’’. This standardized form helped to ensure that the data necessary to understand and guide the entry for new products was collected. It included: product name, source of the nutrition information, i.e. whether it came from the label, website or telephone contact with company, package size and format, preparation instructions for mixes, main ingredients, nutritional composition and NDS entry required to closely match the nutritional content. A sample form is shown in Fig. 2. The format of the MFB also progressed so that it could be updated relatively quickly and easily. The final version was developed using two Microsoft Office programs, Access and Excel. Access was used to assemble and organize the product database
S90
C. Greenberg et al. / Journal of Food Composition and Analysis 22S (2009) S88–S92
Fig. 2. Missing food form.
and data entry rules. The specific product name, including flavour/variety if applicable, was listed along with the NDS entry for a particular serving size. The database also included common package and unit sizes for each item to guide entry when the quantity consumed differed from the reference serving size.
Data entry rules included guidelines for situations when descriptions and/or portions were not specified on the food record, raw to cooked conversions of vegetables and fruits, average sizes and weights of pasta and fresh produce, entering sushi and ribs and calculating volumes of soup. The database
C. Greenberg et al. / Journal of Food Composition and Analysis 22S (2009) S88–S92
S91
Table 1 Mean nutrient values for initial and repeat entry. Overall
Mean initial entry (N = 79)
Mean repeat entry (N = 79)
P-value of differencea
Energy (kJ) Total fat (g) Total protein (g) Total carbohydrate (g)
6774 (1581) 49.0 (20.8) 66.5 (13.7) 227.7 (63.9)
6769 (1530) 48.8 (20.7) 65.0 (13.4) 230.4 (61.0)
0.91 0.73 0.01 0.10
Group = control
Mean initial entry (N = 38)
Mean repeat entry (N = 38)
P-value of differencea
Energy (kJ) Total fat (g) Total protein (g) Total carbohydrate (g)
7288 (1556) 63.5 (18.7) 66.8 (13.2) 223.1 (52.8)
7306 (1566) 63.2 (19.1) 66.0 (13.5) 226.6 (51.9)
0.78 0.79 0.36 0.07
Group = intervention
Mean initial entry (N = 41)
Mean repeat entry (N = 41)
P-value of differencea
Energy (kJ) Total fat (g) Total protein (g) Total carbohydrate (g)
6299 (1468) 35.6 (11.8) 66.3 (14.4) 232.1 (73.0)
6271 (1329) 35.4 (10.8) 64.1 (13.3) 233.9 (68.8)
0.68 0.84 0.009 0.48
a
P-value from one sample t-test.
was exported to Excel for distribution to staff on a quarterly basis. 2.2.3. Review and evaluation of nutrient data entry A review process was established from the outset of the trial to detect nutrient data entry errors. Time and cost constraints made it impossible to check every record that was analyzed so criteria were established to identify records where errors might be more likely to occur. The nutrient data manager checked records when daily energy intake was less than or equal to 800 cal, greater than
or equal to 2800 cal or total fat intake was greater than or equal to 110 g. These limits correspond to about three standard deviations from the mean for energy and fat which was consistent with the criteria used in the Multiple Risk Factor Intervention Trial (Tillotson et al., 1986). The impact on workload for the nutrient data manager of the limits established for ‘‘out of range’’ records was that 10–12% food records needed to be checked and 35% of the records reviewed were changed. To evaluate consistency in nutrient analysis between dietitians two studies were conducted where food records were re-entered
Fig. 3. Repeated entry with initial entry.
S92
C. Greenberg et al. / Journal of Food Composition and Analysis 22S (2009) S88–S92
by a different dietitian. In 1994, 140 randomly selected food records, which represented 10% of the total food records analyzed by that time, were re-entered. There were no differences in the estimates of macronutrients expressed in grams or percent of energy between the first and second analysis when expressed as percent difference. Correlations between all macronutrients were high (mean correlation coefficient = 0.93) (Martin et al., 1994) In 2002 a second re-entry study was conducted to determine the reliability of data entry using different versions of NDS. As mentioned earlier, after 1998 updated versions of NDS were not used so it was important to determine the reliability of entry using early (2.1–2.7) and a later Version (2.91) of NDS. A total of 79 randomly selected 3-day food record sets (38 sets from the control group, 41 sets from the intervention group) that were previously entered in the earlier versions were re-entered in Version 2.91. The subject’s name, study group and study centre were concealed on the food record and distributed to all available dietitians. The results of the second double entry experiment are presented in Table 1 and Fig. 3. There were no significant differences in the means between the initial and repeat entry for energy and total fat and carbohydrate, overall and in the intervention and control groups. There was a very small but statistically significant difference in the grams of total protein between the initial and repeat entry, overall and in the intervention group. Correlations between initial and repeat entries were all greater than 0.9 for the overall, and were similar in the intervention and control groups. This shows strong agreement in nutrient data when food records are entered by different dietitians and different versions of the NDS database and provides convincing evidence that our nutrient analysis system is highly reliable. 3. Discussion and conclusions The nutrient data for the Diet and Breast Cancer Prevention Study are derived from 3-day food records collected from a large
number of subjects over many years. Limitations for all nutritional assessment techniques are well recognized but incorporating quality control procedures for nutrient data entry into study procedures minimized some of the measurement error associated with food records (Prentice, 1996). As a result, high quality nutrient data are available for examination of the relationship between diet and breast cancer risk and phenotype. Designing an effective quality control system for a long-term, multi-centre trial with a large staff needs to be multi-faceted and continual to cope with the magnitude of the task. Selecting highly qualified professionals, providing them with comprehensive training and continuing education, implementing on-going monitoring activities and developing strategies to respond to problems in a timely manner resulted in extremely high quality performance. Furthermore, these processes could easily be applied to single-centre or smaller trials that have similar concerns regarding achieving and maintaining quality data. References Bingham, S.A., Luben, R., Welch, A., Wareham, N., Khaw, K.-T., Day, N., 2003. Are imprecise methods obscuring a relation between fat and breast cancer? Lancet 362, 212–214. Boyd, N.F., Fishell, E., Jong, R., MacDonald, J.C., Sparrow, R.K., Simor, I.S., Kriukov, V., Lockwood, G., Tritchler, D., 1995. Mammographic densities as a criterion for entry to a clinical trial of breast cancer prevention. Br. J. Cancer 72, 476– 479. Freedman, L.S., Potischman, N., Kipnis, V., Midthune, D., Schatzkin, A., Thompson, F.E., Troiano, R.P., Prentice, R., Patterson, R., Carroll, R., Subar, A.F., 2006. A comparison of two dietary instruments for evaluating the fat–breast cancer relationship. Int. J. Epidemiol. 35, 1011–1021. Martin, L.J., Cousins, M.L., Gougeon, L.L., Kruikov, V., Lockwood, G.A., Boyd, N.F., 1994. Quality control of nutrient analysis of food records in a low fat intervention trial. Am. J. Clin. Nutr. 59, 287S. Prentice, R.L., 1996. Measurement error and results from analytic epidemiology: dietary fat and breast cancer. J. Natl. Cancer Inst. 88, 1738–1747. Tillotson, J.L., Gorder, D.D., Duchene, A.G., Grambsch, P.V., Wenz, J., 1986. Quality control in the multiple risk factor intervention trial nutrition modality. Control Clin. Trials 7, 66S–90S.