Nutritional Epidemiology and Dietary Assessment for Patients With Kidney Disease: A Primer

Abstract

Nutritional epidemiology seeks to understand nutritional determinants of disease in human populations using experimental and observational study designs. Though randomized controlled trials provide the strongest evidence of causality, the expense and difficulty of sustaining adherence to dietary interventions are substantial barriers to investigating dietary determinants of kidney disease. Therefore, nutritional epidemiology commonly employs observational study designs, particularly prospective cohort studies, to investigate long-term associations between dietary exposures and kidney disease. Due to the covarying nature and synergistic effects of dietary components, holistic characterizations of dietary exposures that simultaneously consider patterns of foods and nutrients regularly consumed are generally more relevant to disease etiology than single nutrients or foods. Dietary intakes have traditionally been self-reported and are subject to bias. Statistical methods including energy adjustment and regression calibration can reduce random and systematic measurement errors associated with self-reported diet. Novel approaches that assess diet more objectively are gaining popularity but have not yet fully replaced self-report and require refinement and validation in populations with chronic kidney disease. More accurate and frequent diet assessment in existing and future studies will yield evidence to better personalize dietary recommendations for the prevention and treatment of kidney disease.

Introduction

Nutritional epidemiology seeks to understand nutritional determinants of disease in human populations. Some of the earliest epidemiological studies in nutrition investigated the role of diet in the development of cardiovascular disease. Since then, links between diet and numerous chronic diseases, including chronic kidney disease (CKD), have been discovered through epidemiologic investigations. With the rising global prevalence CKD,¹ improved understanding of how diet affects the risk and progression of CKD is imperative to inform care and prevention recommendations.

Epidemiological studies have traditionally relied on self-reported dietary intakes to estimate associations with CKD. However, innovations that overcome limitations of such methods are gaining popularity. This primer offers an overview of nutritional epidemiology, including study designs, exposure definition, traditional and novel diet assessment methods, sources of error, and statistical considerations, to guide researchers and clinicians in interpreting and conducting investigations of diet in CKD.

Study Designs in Nutritional Epidemiology

Randomized controlled trials (RCTs) are considered the most rigorous design to establish causality. Randomized allocation distributes confounding factors similarly between groups and removes the exposure decision from participants and investigators, thereby minimizing confounding. Nutrition RCTs include controlled feeding studies, single nutrient or dietary component studies, and dietary counseling studies (Table 1). One of the largest and longest nutrition RCTs in kidney disease research, the Modification of Diet in Renal Disease (MDRD) Study, investigated the effect of restricting protein intake on glomerular filtration rate decline.² Others have investigated the effect of dietary patterns or nutrients on kidney function decline, blood pressure, anthropometrics, and biochemical derangements.^3,4

Table 1.

Study designs commonly used in nutritional studies of kidney disease

	Description	Strengths	Limitations
Randomized trials
Controlled feeding studies	Study menu is designed to meet desired intake targets All foods and beverages are provided for participants	Nutritional composition of diet can be determined with high accuracy High control over dietary and non-dietary confounders Useful to test efficacy of diet to affect clinical outcomes	Expensive Requires skilled staff and metabolic kitchen to prepare diets High participant burden
Single nutrient or dietary component studies	Only the food(s) or nutrient (supplement) being studied is directly manipulated Instruction regarding other dietary choices may or may not be provided	Effect of an intervention is observed within realistic context of participants’ usual diets	Heterogeneity in intervention effects may result from participant differences in how they incorporate study foods into their habitual diets (e.g. substitution vs. addition) Diet composition is unknown
Dietary counseling studies	Participants are instructed to modify their diets Food is usually not provided	Feasibility observed in real-world setting	Skilled practitioners are required to deliver participant education Diet composition unknown
Observational designs
Prospective cohort studies	Observed diet is associated with subsequent health outcomes	Self-selected diet is observed as it naturally occurs Long follow-up permits observation of hard endpoints (e.g. mortality, incident disease)	Diet is usually only assessed at baseline or early in follow-up and does not capture changes that may occur over time
Cross-sectional studies	Observed diet is associated with concurrent health status	Self-selected diet is observed as it naturally occurs	Direction of associations cannot be determined; possible reverse causation

In human nutrition RCTs, the identity of the intervention is difficult to hide from participants, prohibiting double masking. Systematic errors may arise if intervention adherence and outcome ascertainment differ by randomization group due to participants being cognizant of their group assignment. In addition, a truly unexposed “control” group is impossible in many dietary studies, as all people consume food. Failure to detect an intervention effect may result from insufficient difference between the (lesser exposed) control group and the (more exposed) intervention group, rather than a true lack of effect. Assessment of baseline intake or nutritional status and background diet (the dietary context in which a food or nutrient is consumed) throughout the study duration may be useful to interpret effects. Finally, the expense and difficulty of sustaining adherence to dietary interventions limit the duration and scale of RCTs in nutrition research. Defining “adherence” a priori, and incorporating methods to assess it (e.g. direct observation, return of unconsumed items, biomarker assessment, self-report), is essential to monitor whether the intended intervention is delivered. Ensuring palatability and acceptability of study diets (e.g. by previewing menus or tasting study foods), and communicating clearly and regularly with participants throughout the study duration, can promote retention and adherence. Incorporating a run-in period may be useful to evaluate adherence prior to randomization and standardize baseline exposure. Additional considerations for designing and conducting human nutrition RCTs are detailed elsewhere.^5,6

Due to the long lead times and large sample sizes required to detect intervention effects on hard endpoints, outcomes of nutrition RCTs are commonly intermediate markers of disease risk, with the notable exception of the Prevención con Dieta Mediterránea (PREDIMED) study, a multicenter randomized trial that demonstrated a protective effect of a Mediterranean diet on incident cardiovascular disease, compared to a lower-fat diet.⁷ Trials in kidney disease patients have used declines in glomerular filtration rate of varying magnitudes or the onset of proteinuria as surrogate outcomes for the development of end-stage kidney disease (ESKD)^8,9 which may take years to occur. While sample size and study duration for RCTs investigating hard endpoints (ESKD, death, cardiovascular events) could be minimized by enrolling participants with advanced CKD, dietary interventions may not be equally effective at later versus early stages.¹⁰

Evidence regarding long-term effects of dietary exposures on hard endpoints have mostly originated from observational study designs, predominantly prospective cohort studies, based on participants’ self-reported dietary intakes. Several large cohorts established throughout the past 50 years, including the Chronic Renal Insufficiency Cohort (CRIC) study of adults with existing CKD¹¹ and the Chronic Kidney Disease in Children (CKiD) study of children with existing CKD,¹² continue to yield important discoveries of associations between diet and CKD progression, as well as other important clinical outcomes.^13,14 Cohorts of individuals without CKD at baseline, such as the Atherosclerosis Risk in Communities (ARIC) study, have been used to study the association between diet and incident CKD.¹⁵ In prospective cohort studies, assessment of dietary exposures before outcomes occur supports causal inference. In contrast, cross-sectional studies such as the National Health and Nutrition Examination Survey (NHANES) assess exposure and disease simultaneously and, therefore, yield weaker evidence regarding causal effects of diet on disease but are useful to describe dietary intakes and quantify the burden of insufficient or excess intakes in a population. In observational studies, exposure is selected by participants rather than assigned, and so the influence of confounding factors on observed associations must be accounted for in statistical analyses.

Defining Dietary Exposures

Diet is a complex exposure that varies in composition and quantity within and between days and seasons. The covarying nature of dietary components further complicates definition and statistical modeling of dietary exposures. Conceptualizations of dietary exposures range from a reductionist focus on nutrients to holistic characterizations of dietary patterns (Figure 1).

Figure 1.

Conceptualizations of dietary exposures.

Nutrients are contained within foods and beverages. Foods and beverages with similar nutrient profiles may be grouped together. Dietary patterns encompass the quantity, variety, and combinations of foods and beverages habitually consumed.

Icons were obtained from the Servier Medical Art collection (smart.servier.com, licensed under CC BY 3.0).

Traditionally, single nutrients have been the focus of nutritional epidemiology. Given the kidneys’ key role in metabolizing and excreting several nutrients, they remain important exposures of interest in kidney disease research. Guidelines for CKD management have emphasized dietary protein, phosphorus, potassium, and sodium restrictions.^16–21 However, nutrients are rarely consumed in isolation, except as supplements. Instead, people consume foods, which contain various nutrient and non-nutrient components that may affect health, and the bioavailability of these components may differ depending on the food source and processing. For example, phosphorus from plant sources is less bioavailable than from animal sources, while almost all (>90%) phosphorus from additives is absorbed.²² Furthermore, foods are consumed in combinations that may produce synergistic health effects.²³ Therefore, dietary patterns, which broadly encompass the quantity, variety, and combinations of foods and beverages habitually consumed, may predict long-term health better than single foods or nutrients²⁴ and are the basis of modern dietary guidance.^25,26

Dietary patterns may be defined a priori using predefined criteria (Table 2). For instance, higher adherence to a Mediterranean diet,²⁷ Dietary Approaches to Stop Hypertension (DASH)-style diet,²⁸ and the Dietary Guidelines for Americans²⁹ have been associated with lower risk of incident CKD.^30,31 Alternatively, empirical approaches to defining dietary patterns use statistical methods, such as factor or cluster analysis, that identify combinations of foods consumed in a population.³² Other holistic characterizations of dietary exposures may be defined based on nutrient and food sources or their effects on hypothesized mediators of disease risk.

Table 2.

Examples of a priori dietary patterns and other holistic dietary exposures

Exposure	Definition	References
Alternate Mediterranean Diet Score (aMed)	Scores relative adherence to a Mediterranean-style diet; adapted for use in U.S. populations	27
DASH Diet Score	Scores relative adherence to the blood pressure-lowering dietary pattern tested in the Dietary Approaches to Stop Hypertension clinical trials	28
Healthy Eating Index (HEI)	Scores alignment with the Dietary Guidelines for Americans, with multiple versions corresponding to guideline updates every 5 years	29
Alternative Healthy Eating Index (AHEI)	Scores relative adherence to dietary recommendations that are predictive of chronic disease risk; updated in 2010 based on evolving scientific evidence	33,34
Dietary diversity scores	Assigns greater value to diets with more variety of foods or nutrients	35
Plant-based diet indices	Scores relative adherence to diets richer in plant-derived foods and lower in animal-derived foods, with variations that additionally consider the nutritional quality of plant-derived foods	36
Dietary inflammation index (DII)	Summarizes the inflammatory potential of a diet based on a predefined list of foods, nutrients, and phytochemicals	37
Dietary acid load	Summarizes the balance between acid- and base-producing foods; estimated by calculating potential renal acid load (PRAL) or net endogenous acid production (NEAP) based on dietary protein and mineral intakes	38–40
Processing classification systems	Categorizes foods according to the extent and purpose with which they are processed	41,42

As dietary patterns simultaneously represent multiple foods and nutrients, observed associations between dietary patterns and health cannot be attributed to any single food or nutrient (e.g., protein, sodium, potassium, phosphorus). Thus, holistic dietary patterns complement but do not replace reductionist single nutrient or food studies. These complementary approaches are both valuable for studying diet-disease relationships. [Table 2^{27–29,33–42}]

Diet Assessment Methods

Several methods are available to assess dietary intakes (Table 3). We broadly classify these methods as traditional (self-report) and novel approaches. Each method offers strengths and limitations to consider when selecting the optimal assessment for a given population, study design, and exposure of interest.

Table 3.

Comparison of dietary assessment methods

Method	Information Collected	Strengths	Limitations
Food record	Detailed list of identity, preparation methods, portion sizes, timing, and context of foods consumed on one or multiple (usually consecutive) days	Open-ended Recorded in real-time, avoiding reliance on memory Portion size is measured rather than estimated	Intake may be altered by the act of recording (reactivity) High respondent burden Literacy required Interpreting and entering responses into nutrient analysis software imposes high clinician/researcher burden and expense
24-hour recall	Detailed list of identity, preparation methods, portion sizes, timing, and context of foods consumed in a recent 24-hour period	Open-ended Literacy not required*	Reliance on specific memory - omissions (non-reporting of foods actually consumed) and intrusions (reporting of foods not actually consumed) can introduce bias Trained interviewer required*
Food frequency questionnaire	Frequency and sometimes portion size of foods habitually consumed over a defined time period, usually the previous month or year	Low respondent burden Inexpensive Can capture usual intake of routinely and episodically consumed foods Computerized scanning of paper-based FFQs and online-administered FFQs automate data entry, with direct linkage to databases for nutrient analysis	Reliance on generic memory – cognitively challenging to recall and estimate average intakes over time Lacks detail regarding preparation methods, timing, and context of intake Fixed list does not comprehensively capture foods consumed Not appropriate to estimate absolute nutrient or energy intakes Literacy required, unless interviewer administered
Diet screeners	Limited information about the frequency of food consumption or eating behaviors relevant to a specific clinical or research interest	Quickly and easily completed	Lacks detail – does not comprehensively assess dietary intake
Image-assisted	Photographic record of foods consumed in a single day	Reduces reliance on memory Portion size estimation by software or trained human rater may improve accuracy over self-report	Hidden ingredients and cooking methods may not be discernible in photographed foods Active approaches rely on user to remember to photograph all eating episodes Images captured passively by wearable cameras may be of insufficient quality to accurately identify foods
Wearable technologies	Occurrence, timing, rate, and duration of eating episodes	Minimal user burden Low risk of bias	Does not capture types or portions of foods consumed
Retail sales and purchasing data	Household food purchasing or retail sales data over a defined time period	Minimal or no burden to the consumer	Does not measure actual food or nutrient consumption Expensive to obtain data Details of data collection are not fully disclosed by private companies
Biomarkers	Presence and concentration of compounds in biological specimens, which are indicative of nutrients and foods consumed	Unbiased	Expensive to collect and measure Biospecimens collection can be burdensome to participant (e.g. 24-hour urine) Few validated biomarkers of dietary intake

^*Self-administered online version avoids need for interviewer but requires participant literacy

Traditional Approaches

Food frequency questionnaires.

Food frequency questionnaires (FFQs) query habitual intake over a specified time period using a fixed list of foods, beverages, and possibly nutritional supplements.^24,43 Development or validation of FFQs in the target population is recommended to ensure included foods capture important nutrient sources, which may differ between social, cultural, and ethnic groups. A short 49-item FFQ developed and validated in patients with CKD in France acceptably ranked participants’ intakes of nutrients including phosphorus, potassium, and sodium.⁴⁴ An analogous 57-item version later developed in the US demonstrated acceptable assessment of diet quality but does not assess nutrient intakes.⁴⁵ Similarly, FFQs have been developed or adapted for patients on dialysis.^46–49

Food records.

Food records, or food diaries, require respondents to log dietary intakes in real-time over a specified time period, usually 3 to 7 days. Weighted food records are regarded as the “gold standard” quantitative self-reported diet assessment method and are used to validate other diet assessment methods. KDOQI guidelines recommend use of 3-day food records to assess dietary intakes in clinical practice.¹⁹ Validated online or smartphone application-based records can facilitate real-time data collection, engage users with prompts to record intake, and save time and costs associated with data entry into nutrient analysis software.^50,51

24-Hour Dietary Recalls.

A 24-hour dietary recall queries intake over the past day, either midnight to midnight or the 24-hour period immediately preceding the recall. Ideally, recalls are unanticipated by respondents, so that dietary intakes preceding the recall are not altered. Multiple recalls estimate absolute dietary intakes more precisely than FFQs,⁵² and the accuracy of reporting can be optimized with a multiple-pass interviewing technique.⁵³ While traditionally administered by trained interviewers, an automated online version developed by the National Cancer Institute may be a feasible, less resource-intensive option even for large epidemiological studies.^54,55

Diet screeners.

Screeners, or checklists, query intake of a select few items or eating behaviors (e.g. sodium⁵⁶ or fruit and vegetable⁵⁷ intake). The lack of detail and susceptibility to systematic error limits their utility as a standalone diet assessment method for research studies.

Novel Approaches

Image-assisted dietary assessment.

Image-assisted dietary assessments supplement or replace traditional self-report methods with photos or videos of eating episodes in order to improve objectivity and avoid reliance on memory.⁵⁸ Active image-assisted approaches require participants to photograph foods before and after consumption (to assess net intake) using a handheld device, with a reference object in the frame to assist with portion size estimation. Automated software or human raters subsequently identify foods and portions consumed to estimate nutrient intakes. Comparable estimates of total energy intake were obtained by adults with type 2 diabetes using image-based food records versus weighted food records,⁵⁹ and a mobile phone-based automated image analysis application estimated meal carbohydrate content more accurately than patients with type 1 diabetes.⁶⁰

Passive approaches use wearable cameras to capture daily activities, including eating episodes, without users’ conscious participation. Wearable cameras have been used to identify intakes of forgotten foods on self-reported recalls,⁶¹ and to observe cooking and eating behaviors⁶² in free-living settings.

Wearable technologies.

Wearable technologies offer the ability to passively detect eating events in real-time by monitoring chewing, swallowing, or jaw movements or by monitoring arm gestures associated with eating. These technologies can be used to capture the occurrence, timing, rate, and duration of eating. Bite counters have primarily been used to support weight loss interventions.^63,64 Oral and epidermal sensors are also in development to detect specific nutrients (e.g. sodium, alcohol) in saliva and sweat.⁶⁵ The primary advantage of such technologies is minimization of recall bias. However, as no single sensor is yet able to identify specific foods consumed, their portions sizes, or nutritional composition, they are currently most useful for studying eating behaviors rather than assessing diet composition.

Retail sales and purchasing data.

Market research companies collect point-of-sale data from food retailers, as well as individual- and household-level purchasing data using barcode scanning.⁶⁶ Neither retail sales nor purchasing data directly measure food consumption, as they do not account for culinary preparation, distribution within and beyond the household, and waste. Still, these data may be useful to evaluate the nutritional quality of foods purchased, assess trends over time, and model predicted effects of food reformulation.⁶⁷ For instance, retail sales were used to identify phosphorus-based additives in frequently purchased grocery items.⁶⁸

Dietary intake biomarkers.

Dietary intake biomarkers enable objective measurement of intake based on concentrations of compounds measured in a biospecimen, commonly urine or blood. They are broadly classified as recovery, concentration, and predictive biomarkers. Recovery biomarker concentrations are considered gold-standard approaches for dietary assessment and directly reflect dietary intake.⁶⁹ Only four recovery biomarkers have been identified: 24-hour urinary nitrogen, reflecting protein intake; 24-hour urinary potassium and sodium, reflecting intakes of these minerals; and doubly-labeled water, a measure of energy expenditure that is interpreted as a marker of energy intake in weight-stable individuals.⁷⁰ Concentration biomarkers, such as plasma carotenoids and vitamin C, correlate with dietary intakes but are affected by individual-level characteristics (e.g. adiposity), such that measured concentrations do not directly reflect amounts consumed.⁶⁹ Predictive biomarkers exhibit a time-dependent, dose-response relationship with dietary intakes. They more closely reflect dietary intake than concentration biomarkers but are not completely recovered in biospecimens, like recovery biomarkers. Only 24-hour urinary fructose and sucrose have been identified as predictive biomarkers, and are useful measures in research settings.⁷¹

Dietary biomarkers have traditionally been identified using a hypothesis-driven approach. In recent years, however, the discovery of novel dietary biomarkers has accelerated with the advancement of high-throughput platforms capable of simultaneously identifying many metabolites and proteins in biospecimens. Characterization of the food metabolome and proteome – the sum of compounds resulting from digestion, absorption, and metabolism of consumed foods⁷² – is currently being explored to identify objective biomarker “signatures” of foods and dietary patterns, and to study relationships between diet and CKD. While few foods are uniquely identified by biomarkers, a profile of biomarkers may serve to characterize adherence to a particular dietary pattern. Plasma metabolic signatures of healthy dietary patterns (e.g. AHEI-2010, DASH, Mediterranean) have been identified in adults with CKD,⁷³ and serum metabolomic signatures of plant-based diets have been shown to predict incident CKD.⁷⁴ Serum metabolomic markers of dietary acid load have also been identified⁷⁵ and predict incident CKD.⁷⁵

Candidate biomarkers are primarily identified through cross-sectional correlations with self-reported diet, though controlled feeding studies are needed before biomarkers can serve as objective and quantitative dietary intake markers.⁷⁶ Challenges include differentiating between exogenous (originating from diet) versus endogenous (originating from host or gut microbiota) compounds, and evaluating dose-response and temporal response of compounds to food intake. As most food-derived compounds are only present in urine and plasma acutely after ingestion,⁷⁷ habitual diet assessment may require multiple biospecimen collections over time⁷⁶ or validation of long-term biomarkers from alternative biospecimens (e.g. hair and nails). Finally, most nutritional biomarkers have not been validated in patients with CKD.⁷⁸ Alternative markers or methods may be needed when urinary output is altered or to account for water-soluble solute removal in dialysate.^78,79

Summary of Dietary Assessments

Errors associated with self-reported diet, their impacts on associations with outcomes, and strategies to reduce them are well researched.^80,81 While novel dietary assessment approaches reduce some of these errors, they have not yet replaced traditional methods. Combining self-report with novel assessments may more comprehensively and accurately ascertain dietary intakes.

Sources of Error

Measurement Error: Random

True dietary intakes vary within people, day to day. The degree of intraindividual variability differs by food and nutrient, with higher variability for episodically consumed foods (e.g. organ meats) and nutrients concentrated in few food sources (e.g. vitamin A). A single day’s intake – even if recorded correctly – does not accurately represent an individual’s usual dietary intake of most foods or nutrients. As usual intake is generally more relevant to CKD, such daily deviations from the true long-term average are regarded as random measurement error. Averaging intakes across multiple dietary assessements can improve precision of an individual’s estimated usual intake by reducing random error.⁸² However, the representativeness of collected days should be considered. Dietary intakes may vary across seasons and days of the week,⁸³ and differ on dialysis versus non-dialysis treatment days.^84,85 In addition, dietary intakes on consecutive days are more closely correlated than non-consecutive days, resulting in underestimation of within-person variation in intakes. Gains in precision and representativeness must be weighed against the expense and participant burden of multiple assessments. Distributing shorter automated dietary assessments (e.g. automated self-administered 24-h recalls^54,86) over time, and including weekdays vs. weekends and dialysis vs. non-dialysis days, may help maximize representativeness while reducing participant burden and expense.

For epidemiological research questions, group-level (rather than individual) usual intake estimates are typically of interest. On any given day, observed intakes in a population will include individuals with exceptionally high and low true single-day intakes, but the distribution will cluster around a central mean. Thus, a single 24-hour recall or record per person is considered sufficient to estimate population mean intakes, but extreme high and low consumers will flatten the estimated distribution of intakes and overestimate the proportion of a population in the distribution tails. As such, comparisons of population intakes to some target – for instance, an estimated minimum requirement – will overestimate the proportion of deficient individuals. Statistical methods are available to more precisely estimate the usual intake distribution in a population with as few as two recalls obtained from a subset of respondents.^87–89 Such methods should be utilized when assessing the adequacy or insufficiency of population intakes relative to a recommended target, or when comparing usual intakes between two or more groups.

In contrast to 24-hour recalls and food records, FFQs solicit average consumption over a defined time period and are intended to capture usual intake. However, dietary intakes may change over extended periods of follow-up, especially in persons with CKD, who may experience changes in appetite and receive dietary counseling. If only baseline diet is assessed, unmeasured subsequent shifts in intakes could result in misclassification of exposure status and attenuation of estimated associations with CKD. Repeating assessments and averaging intakes across repeated FFQs may reduce random measurement error arising from intraindividual changes in dietary intakes over time.⁹⁰

Measurement Error: Systematic

While repeating dietary assessments can improve precision by reducing random measurement error, it will not address systematic misreporting of intakes. The degree of misreporting varies for different population groups, foods, and dietary assessment tools.^80,91 For instance, people may overreport intakes of “healthy” foods and underreport “unhealthy” foods to align with social norms (social desirability and social approval biases),⁹² and people with low true intakes tend to overreport their intake, while those with truly high intakes tend to underreport (the “flattened slope phenomenon”).⁸¹ Regression calibration can help correct risk estimates for biased measurements of self-reported dietary exposures when a less biased reference assessment is available for at least a subset of the study population.^81,93

Food Composition Databases

Whether dietary intakes are self-reported or observed, conversion to nutrients and foods requires linkage to nutrient databases. Misestimation of nutrient contents may arise due to geographical, seasonal, or other natural sources of variation in the nutritional composition of foods. For instance, selenium can vary substantially due to differences in soil content.⁹⁴ Misestimation also arises from variation in the composition of prepared foods and manufactured items. While some databases include nutrient contents for branded foods or restaurant dishes, actual nutrient contents may differ substantially from reported values.^95,96 In addition, nutrient losses during food preparation or cooking may not be considered. Finally, database values do not account for biological availability of consumed nutrients. For instance, phosphorus intakes from natural and added sources are weighed equally in database summations, though true absorption varies substantially.²² Cautious interpretation of nutrient intakes is warranted, as summed database totals may not fully represent actual consumed or biologically available nutrients.

Statistical Modeling

Energy Adjustment

A methodological consideration unique to nutritional epidemiology is whether and how to adjust for energy intake.^90,97–99 People consuming more energy generally consume greater absolute amounts of nutrients, and energy intake itself may be associated with CKD. Associations between absolute dietary intakes and CKD may, therefore, be confounded by energy intake. Thus, epidemiological analyses of associations between dietary intake and CKD typically aim to statistically isolate between-person variation in intakes due to energy intake from variation explained by changes in dietary composition. Controlling for total energy also helps to reduce error in self-reported dietary intakes, as error in self-reported energy intake measurement is correlated with error in self-reported intakes of nutrients and foods.⁹⁹ The interpretation of regression coefficients from energy adjusted models differs depending on whether nutritional exposures are modeled continuously or categorically¹⁰⁰ and whether other dietary covariates are included (substitution).¹⁰¹

Substitution

When energy balance is maintained, increasing intake of a macronutrient or energy-containing food or beverage requires decreasing intakes of others, and its biological effect may differ depending on what it displaces in the diet. For instance, replacing saturated fat with unsaturated fat is associated with lower coronary heart disease risk whereas no association is observed when it is replaced with refined carbohydrates.¹⁰² Specific isocaloric food or macronutrient substitutions can be modeled by including total energy and all other energy sources excluding the food or macronutrient to be substituted as covariates.¹⁰¹ Modeled substitutions should ideally represent realistic exchanges and consider how foods are actually consumed within a dietary pattern.¹⁰³

Models that only adjust energy are difficult to interpret, as associations may be attributable to increasing intake of the dietary exposure or decreasing intakes of unspecified other energy sources. Substitution analyses better inform dietary guidance by identifying appropriate dietary replacements to reduce CKD risk.¹⁰¹ A clear and specific research question will ultimately guide model specification.

Future directions

Dietary recommendations for chronic disease prevention and management have evolved from single nutrient modifications to a more holistic food-based dietary pattern approach. A similar trajectory is evident in CKD guidelines, which call for research investigating the effects of dietary patterns on clinical outcomes in CKD patients.¹⁹ Refinement and validation of novel dietary assessment methods in CKD populations, including patients on dialysis, will yield more objective measurements of intake, while automated versions of traditional methods offer the potential to assess diet more frequently, and in more detail, than previously possible.

In addition, advancements in “omics” profiling – of the genome, transcriptome, proteome, metabolome, and gut microbiome – offer opportunities to investigate mechanisms by which diet affects kidney health. Genetic and epigenetic variations affect metabolic responses to diet, and diet can alter gene expression, contributing to interindividual transcriptomic, proteomic, and metabolomic variation that may mediate diet-CKD associations.^104,105 Diet can also alter the composition, diversity, and function of gut microbiota, and modulate gut epithelial permeability,¹⁰⁶ thereby altering metabolite generation (e.g. uremic toxins, short-chain fatty acids)¹⁰⁷ and translocation across the intestinal wall.¹⁰⁴ As gut microbiome disturbances may contribute to CKD disease progression and complications,^108,109 its modulation by diet may impact prognosis. Omics technologies are increasingly applied to advance CKD^110,111 and diet research.^104,112 Application of these advancements in new and existing studies, combined with more traditional approaches to risk stratification based on sociodemographic factors, health status, and disease stage, will yield evidence to form more personalized dietary recommendations tailored to patients’ characteristics, social background, etiology and stage of CKD as well as co-existing comorbidities.