Dietary Plant and Animal Protein Intake and Decline in Estimated Glomerular Filtration Rate Among Elderly Women

A 10-Year Longitudinal Cohort Study

Amélie Bernier-Jean; Richard L. Prince; Joshua R. Lewis; Jonathan C. Craig; Jonathan M. Hodgson; Wai H. Lim; Armando Teixeira-Pinto; Germaine Wong


Nephrol Dial Transplant. 2021;36(9):1640-1647. 

In This Article

Materials and Methods

Study Participants

Participants of the Longitudinal Study of Ageing Women cohort were recruited as part of the Calcium Intake Fracture Outcome Study, a 5-year double-blind randomized controlled trial of calcium supplements for the prevention of osteoporotic fractures in older women.[12] In 1998, 24 800 women were randomly selected from the 33 336 women >70 years old on the electoral roll in Western Australia. A total of 5586 (22.5%) responded to the initial invitation. A total of 1222 women were excluded for having a pre-existing metabolic bone disease, receiving bone active agents or having a projected survival of <5 years. In total, 1510 were enrolled in the study. As previously reported, the characteristics of the participants were comparable to the age-matched general population in their medication and their disease burden but had overall higher socioeconomic status.[12] All participants gave written informed consent and the Human Ethics Committee of the University of Western Australia, and the Human Research Ethics Committee of the Western Australian Department of Health approved the study (approval number 2009/24).

At the end of the study, approval was granted for a 5-year extension.[13] This study is based on the data from the baseline, and 5- and 10-year assessments and included all participants who completed the food frequency questionnaire (FFQ) at baseline and had at least one measurement of kidney function throughout the study.

Dietary Assessment

Participants completed a validated semi-quantitative FFQ, the Dietary Questionnaire for Epidemiological Studies version 2,[14,15] at baseline, 5 and 10 years. The questionnaire measured their usual eating and drinking habits over the previous 12 months and included pictures of portion sizes. We excluded questionnaires with implausible energy intakes [<3350 kJ (800 kcal) or >17 575 kJ (4200 kcal)/day].

We estimated the consumed amount of each food item in grams from the frequency of consumption and the portion size. We calculated the protein intake from each food item using the AUStralian Food and NUTrient Database (AUSNUT) 2011–13 food nutrient database.[16] We obtained the total plant protein intake by combining the protein content from fruits, vegetables, beans and legumes, grain foods and nuts, and the total animal protein intake was obtained by combining the protein content from meat, poultry, fish, eggs and dairy products. For items containing ingredients of both plant and animal origin, we divided each item in its constituting ingredients according to the AUSNUT 2011–13 food recipe repertory.[17] We then calculated the protein content of each ingredient and allocated it to either plant or animal origin following the same definitions as above. For quality assessment, we recalculated the total protein intake from the sum of the plant and animal protein intake and compared it with the protein intake previously measured in the same cohort. The intraclass correlation coefficient was 0.997 [95% confidence interval (CI) 0.996–0.997]. We excluded questionnaires with extreme total protein intake (lower or >3.5 SDs from the mean) to avoid generating estimates based on single measurements.

Covariates of Interests

Age, body mass index (BMI), smoking status, physical activity, socioeconomic status, medical history [HTN, prevalent coronary heart disease (CHD), diabetes, cerebrovascular disease, heart failure and peripheral arterial disease] and current medication (antihypertensive agents and statins) were collected at baseline. Participant's socioeconomic statuses were measured using the Socio-Economic Indexes for Areas 1991 following the Australian Bureau of Statistic method.[18] We coded participant's comorbidities according to the International Classification of Primary Care-Plus method.[19] We also adjusted for treatment allocation in the original trial.


The outcomes of interest were the interactions of plant and animal protein intake with time on estimated glomerular filtration rate (eGFR). Serum creatinine and cystatin C were measured at baseline, and 5 and 10 years. Serum creatinine was analyzed using an isotope dilution mass spectrometry traceable Jaffe kinetic assay for creatinine on a Hitachi 917 analyzer (Roche Diagnostics GmbH, Mannheim, Germany). Serum cystatin C was analyzed on the Siemens Dade Behring Nephelometer, traceable to the International Federation of Clinical Chemistry Working Group for Standardization of Serum cystatin C and the Institute for Reference Materials and Measurements certified reference materials. GFR was estimated using the Chronic Kidney Disease Epidemiology Collaboration equation derived from serum creatinine and cystatin C.[20]

Statistical Analyses

Descriptive statistics are presented as means and SDs and absolute and relative frequencies as appropriate. We constructed longitudinal mixed linear models with random intercepts. A random slope could not be tested due to the number of participants with kidney function assessed at two time-point or less (Supplementary data, Table S1). After assessing linearity, we parametrized plant and animal protein intake, age at baseline, BMI, total energy intake and time as continuous linear variables. Plant and animal protein intake, total energy intake and BMI were time-dependent variables in that the eGFR was regressed over the concurrent intake of plant and animal protein at 0, 5 and 10 years. In the multivariable models, plant protein intake was always adjusted for animal protein intake and vice versa. Missing covariates were imputed using multiple imputations with chained equations (Supplementary data, Table S2). The outcome variable, eGFR, was normally distributed.

We conducted a multivariable analysis with the subset of variables, including age, BMI, smoking status, physical activity, socioeconomic status, medical history and current medication, that presented a P < 0.25 in the univariable analyses. The final model was selected using a stepwise backward approach, where we kept variables that were either significantly associated with eGFR at a P < 0.05 or that confounded the effect of the primary explanatory variables. Time, animal, plant protein intake and their interactions were always kept in the model as well as total energy intake for face validity. We also tested for three-way interactions between plant and animal protein intake, time and CKD status at baseline (defined as eGFR <60 mL/min/1.73 m2), diabetes and HTN status (defined as either baseline systolic blood pressure >140 mmHg or diastolic blood pressure >90 mmHg or need for antihypertensive medication). Protein intake from each food group (i.e. fruits and vegetables, grains, legumes and beans, nuts and seeds, meat, poultry, fish, eggs and dairy products) was also assessed. Diagnostic plots showed normally distributed residuals for all models. To assess the robustness of our findings to truncation of follow-up by death, we performed a sensitivity analysis using a fully conditional model where the slopes of eGFR were stratified by survival time (those who have survived to 10 years versus those who have died before 10 years).[21] Significance level was set at 0.05 in two-tailed testing. Plant and animal protein intake for each participant was calculated using IBM SPSS Statistics version 24, and the linear mixed models were constructed using SAS® 9.4 University Edition.