SEARCH

SEARCH BY CITATION

Keywords:

  • allergy;
  • asthma;
  • confounding;
  • diet;
  • effect modification

Abstract

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author contributions
  10. Funding
  11. Ethical approval
  12. Coflict of interest
  13. References

Objective

To propose a comprehensive set of confounders and effect modifiers that should be considered in epidemiologic investigations.

Methods

Two reviewers independently critiqued studies included in a recent systematic review and extracted data on the confounders and effect modifiers that were considered and the approaches used to justify inclusion.

Results

Of the 62 studies reviewed, 20 were cohort, 16 case–control, 25 cross-sectional studies, and one ecologic study. All cohort, cross-sectional, and ecologic studies had some adjustment for confounding or consideration of effect modification, but this was only the case for 7/16 (44%) case–control studies. Of the 53 studies that considered confounding or effect modification, 39/53 (74%) gave no justification for the inclusion of the variables considered. Studies that justified the inclusion of the variables did so based on empirical evidence (n = 10), conceptual justification (n = 7), or a combination of the two (n = 3). Confounding was handled mainly by using regression modeling, but some case–control studies utilized matching and anova. Ten studies handled effect modification by stratification, eight tested for interaction, and five used both strategies.

Conclusions

We have found substantial shortcomings in the handling of confounding and effect modification in studies of diet and development of childhood asthma/allergies. Selection of variables should be based on conceptual considerations and empirical evidence. Using this approach, we have proposed a comprehensive set of confounders and effect modifiers that need to be considered in future studies.


Background

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author contributions
  10. Funding
  11. Ethical approval
  12. Coflict of interest
  13. References

There is a substantial international interest in the etiologic role of nutrients and other dietary factors in the development of asthma and atopic disease. Although observational studies have reported associations between nutrients and asthma and atopic outcomes, there are concerns that these associations may be a consequence of confounding by complex social, behavioral, and lifestyle factors acting across the life course [1]. The scientific and resource consequence of confounding in observational studies of nutrition and disease has been highlighted by the failure of many well-conducted randomized controlled trials (RCTs) to show beneficial effects of nutrient supplementation despite numerous reports of beneficial associations between nutrients and cardiovascular disease, cancer, and all-cause mortality. To achieve the validity in the evidence provided by observational epidemiologic studies, extra effort therefore needs to be dedicated to the identification, measurement, and adjustment of potential confounding variables [2-5]. It is for this reason that the Strengthening The Reporting of Observational Studies in Epidemiology (STROBE) guidelines highlight the importance of considerations of confounding in observational studies [6].

Our recent systematic review found a large body of relevant epidemiologic evidence relating early life and childhood dietary exposures to atopic and asthma outcomes, but much of this was judged to be of poor methodological quality on the grounds of, among other things, the potential for confounding bias [7]. In order to catalyze improvements in the methodological quality and reporting of future epidemiologic investigations, we undertook a further critique of the 62 international studies [8-69] in our dataset to identify and propose a comprehensive set of confounders and effect modifiers that should be considered and suggest ways in which these may be accounted for.

Confounding can be defined as ‘the distortion of a measure of the effect of an exposure on an outcome due to the association of the exposure with other factors that influence the occurrence of the outcome.’ [70] Confounders are variables used to ameliorate the effect of confounding by suitable adjustment [70]. The selection of confounders is guided by the knowledge of the subject matter under investigation [70, 71], and while associated with both outcome and exposure, they are not on the causal pathway between exposure and outcome [71]. Confounding is addressed by appropriate adjustment in the design of studies and, more commonly, statistical analysis and a variety of design-related strategies, which include restriction, matching, stratification, and regression modeling [71].

Effect modification can be defined as the ‘variation in the selected effect measure for the factor under study across levels of another factor’ [70, 71]. Although not a bias, appropriate consideration of effect modification helps to uncover the details of the property of the effect under study [72]. In practice, effect modification is assessed by (i) including the product of the exposure and the effect modifier(s) in a regression model and computing statistical significance and (ii) stratifying the effect of the exposure on the outcome by the effect modifier(s) [71-73].

Methods

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author contributions
  10. Funding
  11. Ethical approval
  12. Coflict of interest
  13. References

Details of the methodology for the identification, selection, and inclusion of the studies have been previously reported [7]. In summary, the included studies [8-69] were of children aged ≤16 years and included mothers during pregnancy. The exposures of interest were food consumption, nutrient intake, and nutrient biomarkers during pregnancy, infancy, and childhood. In addition, we included studies investigating both nutrient deficiency and the role of increased nutrient consumption, whether through diet or supplementation. The primary outcomes of interest were asthma and atopic disorders as assessed by any objective measure, that is, the incidence of physician-confirmed asthma/atopic disease, lung function, and atopic sensitization. Secondary outcomes of interest were measures of disease severity, health economic data, and data on adverse effects.

These studies were reappraised by two independent reviewers (UN and BIN) to extract detailed information on the confounders and effect modifiers included in the studies; examine the approaches and justifications employed in the selection of the confounders and effect modifiers; appraise the analysis strategies utilized to handle the confounders and effect modifiers. A customized data extraction form was used to record the extracted set of information. After the independent review of studies, any disagreements were resolved through discussion between the two reviewers. A third reviewer arbitrated if discussion could not be reached. BIN and GD did not participate in the discussion of the studies in which they were co-authors.

The information extracted from the studies was transferred to a Microsoft Excel spreadsheet, and basic descriptive statistics of study characteristics were computed and stratified by study design.

Results

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author contributions
  10. Funding
  11. Ethical approval
  12. Coflict of interest
  13. References

The main features of the included studies are presented in Table 1, arranged in alphabetical order of authorship. Table 2 contains the summary statistics relating to issues of confounders and effect modifiers, extracted from the information in Table 1.

Table 1. Treatment of confounders and effect modifiers in studies investigating the relationship of diet to asthma and allergic diseases
Reference, countryStudy design and populationExposureOutcomesConfounders (C) and effect modifiers (EM) consideredJustification/approaches to the selection of C and EMApproaches used to deal C and EM
  1. BMI, body mass index; C, confounders; EM, effect modifiers; FFQ, Food Frequency Questionnaire; GDP, gross domestic product; ISAAC, The International Study of Asthma and Allergies in Childhood; RCTs, randomized controlled trials; SPT, skin prick test.

Antova et al., Multicenter [7]Cross-sectional study of 7- to 11-year-old children; n = 20271Intakes of fish, fruits, vegetablesCough and wheeze assessed by symptom-based questionnaire

C: age, sex, area, pets at home, indoor moisture at home, use of gas oven for heating, additional gas heating, smokers in household, maternal education, paternal occupation, parental allergy, questionnaire respondent, overcrowding at home

EM: country of study

No justification given

C: Multiple regression

EM: Interaction and stratification

Arora et al., India [8]

Case–control study of 2- to 12-year-olds

Cases: n = 19

Controls: n = 25

Serum levels of vitamin AAsthma based on documented clinical history

C: age, sex

EM: None

No justification givenC: Matching: cases and controls matched by age and sex
Awasthi et al., India [9]Cross-sectional study of 6- to 7- and 13- to 14-year-olds; n = 6000Fruits and vegetablesISAAC-based wheeze and asthma

C: traffic pollution, parental smoking, cooking fuels, proximity to pets, breastfeeding, TV watching, exercise, antibiotic and paracetamol use in year 1, and parental education

EM: None

No justification givenC: Multiple regression

Batlle et al.,

Mexico [10]

Cross-sectional study of 6- to 7-year-olds; n = 1476Mediterranean dietISAAC-based asthma, wheeze, allergic rhinitis

C: sex, siblings, birth weight, BMI, breastfeeding, parental education, child's exercise, smoking exposure at home, parental allergic history, animals at home during yr 1, living in ranch during pregnancy, premature birth, respiratory infections during early life, silhouette, type of combustible used for cooking at home, sleeping alone at home, and mold dampness at home

EM: gender, smoking, maternal education, maternal asthma and allergic rhinitis

Conceptual: based on previous literature.

Empirical: based on statistical test in relation to the exposure and outcome

C: Multiple regression

EM: Stratification

Bäck et al., Sweden [11]Cohort study of 6-year-old children; n = 123Vitamin D3 intakeISAAC-based asthma, atopic dermatitis, and allergic rhinitis

C: Parental atopy, family allergic history

EM: Family allergic history

Conceptual: family allergic history as an important risk factor for allergy

C: Multiple regression

EM: Interaction and stratification

Burney et al., Multicenter [12]

Case–control study of 20- to 45-year-olds;

Cases: n = 569

Controls: n = 576

Plasma seleniumSelf-reported asthma

C: gender, age, smoking, SES, supplement use, paracetamol use, and BMI

EM: gender, atopy, and smoking

No justification given

C: Multiple regression

EM: Interaction and stratification

Camargo et al., the United States [13]Cohort study of 3-year-olds; n = 1194Vitamin D intakeQuestionnaire-based wheeze and eczema

C: sex, birth weight, income, maternal age, maternal pre-pregnancy BMI, passive smoking exposures, breastfeeding, siblings <12 years, maternal asthma/eczema, paternal asthma/eczema, maternal intake of supplements, calcium, and retinol

EM: pre-pregnancy BMI, maternal intake of calcium and retinol, parental asthma, white maternal race, and season of last menstrual period

No justification given

C: Multiple regression

EM: Interaction and stratification

Castro-Rodriguez et al., Spain [14]Cross-sectional study of preschoolers of 4.08 mean age; n = 1784Mediterranean dietISAAC-based wheeze

C: age, birth weight, livestock during pregnancy, cesarean section, antibiotic during year 1, acetaminophen intake during year 1, rhinoconjunctivitis, dermatitis, paternal asthma, maternal asthma, maternal age, maternal education, paternal smoking, maternal smoking, physical activity, cats at home in the past 12 months

EM: None

Empirical: based on statistical test (P < 0.05) in bivariate association between the covariates and the outcomeMultiple regression
Chatzi et al., Greece [15]Cross-sectional study of 7- to 18-year-olds; n = 690Fruits, vegetables, and Mediterranean dietISAAC-based wheeze and allergic rhinitis: skin prick test (SPT) for atopy

C: age, sex, BMI, parental asthma, number of siblings

EM: None

No justification givenMultiple regression
Chatzi et al., Spain [16]Cross-sectional study of 6.5-year-olds; n = 460Fish, fruits, vegetablesQuestionnaire-based wheeze; SPT

C: gender, parental asthma, parental atopy, maternal smoking, BMI at 6.5 years, parental education and social class, breastfeeding, fish intake during pregnancy, number of siblings

EM: None

Empirical: based on statistical test (P < 0.20) in bivariate association between the covariates and the outcomeMultiple regression
Chatzi et al., Spain [17]Cohort study of 6.5-year-olds; n = 460Mediterranean dietQuestionnaire-based wheeze; SPT

C: gender, maternal age, parental education and social class, maternal smoking during pregnancy, supplement use during pregnancy, parental asthma, breastfeeding, lower tract infection at year 1, birth weight, gestation, birth order, number of siblings, BMI at year 6.5

EM: None

Empirical: based on statistical test (P < 0.20) in bivariate association between the covariates and the outcomeMultiple regression
Cook et al., UK [18]Cross-sectional study of 8- to 11-year-olds; n = 2650Fruits and vitamin C in plasmaQuestionnaire-based wheeze; lung function (FEV1)

C: age, sex, height, town of study, instrument, observer, room temperature, obesity index, social class, salivary concentration of cotinine, and birth weight

EM: None

Conceptual: based on previous literature (only for birth weight and not other confounders)Multiple regression
Devereux et al., UK [19]Cohort study of 5-year-olds; n = 1861Plasma vitamin E and zincISAAC-based asthma, wheeze, eczema, and hayfever; SPT

C: maternal age, maternal atopy, maternal smoking, maternal vitamin intake, maternal vitamin E or zinc intake, paternal social class, maternal age at leaving full-time education, deprivation index, birth weight, birth head circumference, crown-heel length, sex, birth order, breastfeeding, and use of antibiotics in year 1

EM: None

No justification givenMultiple regression
Devereux et al., UK [20]Cohort study of 5-year-olds; n = 1924Plasma seleniumISAAC-based wheeze and asthma

C: gender, maternal atopy, maternal smoking, maternal age, parity, paternal social class, maternal age at full-time education, birth weight, crown-heel length, head circumference, maternal plasma ascorbate, α-tocopherol, zinc, antibiotics during infancy, breastfeeding, and deprivation index

EM: None

Conceptual: based on the previous literature.

Empirical: based on statistical test (P < 0.25) in bivariate association between the covariates and the outcome

Multiple regression
Devereux et al., UK [21]Cohort study of 5-year-olds; n = 1212Vitamin D intakeISAAC-based wheeze and asthma; SPT; bronchodilator response, spirometry, and exhaled nitric oxide

C: maternal atopy, maternal age, maternal smoking, maternal age at leaving full-time education, paternal social class, deprivation index, breastfeeding, sex, use of antibiotics in year 1, birth weight, birth order, season of last menstrual period, maternal intakes of vitamin E, zinc, and calcium

EM: None

No justification givenMultiple regression
Ellwood et al., Multicenter [22]Ecological study of 53 countriesPer capita food intakeISAAC-based asthma allergic rhinoconjunctivitis, and atopic eczema

C: GDP of countries

EM: None

Conceptual: based on the previous literature – GDP correlated with allergic symptomsMixed models regression
El-Kholy et al., Egypt [23]

Case–control study of 2- to 12-year-olds

Cases: 40

Controls: 20

Serum and hair zinc and copperHospital-based bronchial asthma and atopic dermatitis

C: age, sex, socioeconomic status

EM: None

No justification givenMatching: cases and controls matched by age, sex, and socioeconomic status
Erkkola et al., Finland [24]Cohort study of 5-year-olds; n = 1669Vitamin D intakeISAAC-based asthma, allergic rhinitis, and atopic eczema

C: sex, place of birth, gestation, maternal age, maternal education, maternal smoking during pregnancy, siblings, parental asthma and allergic rhinitis, eczema by 6 months, pets at home by 1 year, fruits, vegetables, vitamin C, vitamin E, selenium, zinc

EM: None

No justification givenMultiple regression
Ermis et al., Turkey [25]

Case–control study of children mean age of 7.6 years;

Cases: n = 41

Controls: n = 30

Serum zinc, copper, and magnesium levelsReported bronchial asthma

C: None

EM: None

NANA
Farchi et al., Italy [26]Cross-sectional of 6- to 7-year-olds; n = 4104Vegetables, fruits, fat products, dairy productsISAAC-based wheeze, shortness of breath, and allergic rhinitis

C: sex, study area, paternal education, household crowding, maternal smoking, paternal smoking, dampness or mold in child's room, and parental asthma

EM: None

No justification givenMultiple regression
Fitzsimon et al., Ireland [27]Cohort study of 3-year-olds; n = 631Vegetables, fruits, oily fishQuestionnaire-based asthma

C: maternal age, maternal BMI, maternal smoking, exposure to smoke at home, mold or dampness in the home, pollution problems in local area, maternal university education, GMS status, income, parity, birth weight, gender, breastfeeding

EM: hospital of birth

No justification given

C: Multiple regression

EM: Stratification

Forastiere et al., Italy [28]Cross-sectional study of 6- to 7-year-olds; n = 4104Citrus and kiwi fruitsISAAC-based wheeze, shortness of breath, and allergic rhinitis

C: sex, study area, paternal education, household crowding, maternal smoking, paternal smoking, dampness or mold in the room, parental asthma

EM: sex area of residence, paternal education, parental smoking, parental asthma, life history of asthma

No justification given

C: Multiple regression

EM: Interaction

Gale et al., UK [29]Cohort study up to the age of 9 years; n = 46625-(OH)-vitamin concentrationQuestionnaire-based eczema and asthma

C: maternal education, season of birth

EM: None

No justification givenMultiple regression
Garcia et al., Columbia [30]Cross-sectional study of 6- to 7- and 13- to 14-year-olds; n = 7085Food consumptionISAAC-based wheeze and asthma

C: age, sex, contact with animals, cesarean section, breastfeeding, presence of cat/dog at home in year 1, exercise, TV watching, maternal smoking during year 1, maternal education, frequency of traffic around the house, type of cooking fuel, family size, use of acetaminophen and antibiotics during year 1

EM: None

No justification givenMultiple regression
Garcia-Marcos et al., Spain [31]Cross-sectional study of 6- to 7-year-olds; n = 20106Mediterranean dietISAAC-based asthma and rhinoconjunctivitis

C: sex, obesity, maternal smoking, siblings, exercise

EM: sex

No justification given

C: Multiple regression

EM: Stratification

Gilliland et al., the United States [32]Cross-sectional study of 11- to 19-year-olds; n = 2566Antioxidant vitamins, fruits, and vegetablesLung functions

C: parental education, household income, BMI, age, sex, insurance status, personal smoking, environmental smoke exposure, respiratory illness

EM: sex, asthma status

Empirical: based on causing a% change in estimate (≥10%) of the association between exposure and outcome

C: Multiple regression

EM: Interaction and stratification

Harik-Khan et al., the United States [33]Cross-sectional study of 6- to 17-year-olds; n = 4093Vitamins A, C, and E and carotenoidsQuestionnaire-based asthma

C: age, sex, household size, BMI, household head educational status, household head sex, household head employment, parental allergic history, race, smoker in household

EM: None

No justification givenMultiple regression
Hijazi et al., Saudi Arabia [34]

Case–control study of 12-year-olds

Cases: n = 114

Controls: n = 202

Foods and nutrientsISAAC-based asthma

C: place of residence, family allergic history, SPT positivity, maternal education, sex, social class

EM: None

Conceptual: based on the previous literature.

Empirical: based on the statistical test in relation to the outcome

Multiple regression
Hozyasz et al., Poland [35]

Case–control study of 1- to 9-year-olds

Cases: 25

Controls: 18

Serum α-tocopherolAtopic dermatitis

C: None

EM: None

NANA
Hozyasz et al., Poland [36]

Case–control study of 1- to 9-year-olds

Cases: 19

Controls: 18

Plasma retinolAtopic dermatitis

C: None

EM: None

NANA
Huang et al., Taiwan [37]Cross-sectional study of 13- to 17-year-olds; n = 1166NutrientsQuestionnaire-based asthma and allergic rhinitis

C: sex and urbanization

EM: None

No justification givenMultiple regression
Hyppönen et al., Finland [38]Cohort of 7648 adultsVitamin D supplementationQuestionnaire-based asthma and allergic rhinitis; SPT

C: family allergic history and asthma, social class, maternal attitude to health education, maternal frame of mind during delivery, pregnancy wanted by mother, sex, birth order, maternal age, maternal education

EM: None

No justification givenMultiple regression
Kalayci et al., Turkey [39]

Case–control study of 13- to 15-year-olds

Cases: 14

Controls: 12

Serum α-tocopherol, beta carotene, and ascorbic acidHospital-based asthma

C: None

EM: None

NANA
Kocabaş et al., Turkey [40]

Case–control study of 0- to 3-year-olds

Cases: 61

Controls: 61

Serum seleniumQuestionnaire-based wheeze

C: None

EM: None

NANA
Lewis et al., UK [41]Cross-sectional study of 4- to 6-year-olds; n = 1162FruitsISAAC-based wheeze and asthma

C: age, sex, area, smokers at home, distance from roads, Townsend index

EM: None

No justification givenMultiple regression
Litonjua et al., the United States [42]

Cohort study of 2-year-olds;

n = 1290

Antioxidant nutrients and mineralsQuestionnaire-based wheeze and eczema

C: birth weight, sex, maternal age, maternal pre-pregnancy BMI, breastfeeding, siblings ≤12 years, postnatal passive smoking exposure, family income, parental asthma, parental eczema, maternal intake of fruits and vegetables, and child multivitamin intake in year 1

EM: None

Empirical: based on statistical test (P < 0.05) in relation to the outcome and% of change (8%) in the association between the exposure and outcomeMultiple regression
Martindale et al., UK [43]Cohort study of 2-year-olds; n = 1924AntioxidantsISAAC-based wheeze and eczema

C: sex, maternal age, paternal social class, maternal atopy, maternal smoking, siblings, antibiotic use, vitamin E or C

EM: None

No justification givenMultiple regression
McKeever et al., the United States [44]Cross-sectional study of 6- to 16- (n = 4428) and 17- to 59- (n = 5858) year-oldsMarkers of antioxidants, lipids, and mineralsSPT

C: age, sex, smoking, BMI, poverty index, race, survey design

EM: None

No justification givenMultiple regression
Milner et al., the United States [45]Cohort study of 3-year-olds; n > 8000Multivitamin supplementationQuestionnaire-based asthma and food allergies

C: education, race, sex, breastfeeding, degree of prematurity, household smoking, household income, enrollment in day care

EM: race, breastfeeding

Empirical: based on statistical test (P < 0.05) in relation to the outcome

C: Multiple regression

EM: Stratification

Miyake et al., Japan [46]Cohort study of 16- to 24-month-olds; n = 763Vegetables, fruits, and antioxidantsISAAC-based wheeze and eczema

C: maternal age, gestation, residential municipality, family income, parental education, parental asthma, allergic rhinitis, and atopic eczema, maternal diet during past 1 month, season of data collection, maternal smoking during pregnancy, older siblings, sex, birth weight, household smoking, breastfeeding, age

EM: None

No justification givenMultiple regression
Miyake et al., Japan [47]Cohort study of 16- to 24-month-olds; n = 763Dairy products, calcium, and vitamin DISAAC-based wheeze and eczema

C: maternal age, gestation, residential municipality, family income, parental education, parental asthma, allergic rhinitis, and atopic eczema, maternal diet during past 1 month, season of data collection, maternal smoking during pregnancy, older siblings, sex, birth weight, household smoking, breastfeeding, age

EM: None

No justification givenMultiple regression
Mizuno et al., Japan [48]

Case–control study

Cases: n = 26 (mean age 5.5 years)

Controls: n = 25 (mean age 6.4 years)

Serum vitamin AAsthma

C: None

EM: None

NANA
Murray et al., UK [49]

Nested case–control study of 3- to 5-year-olds

Cases: n = 37

Controls: n = 37

Multivitamins, vitamins, minerals, and fatty acidsWheeze

C: total energy, total fat (matching factors), sex, month of birth, pet ownership, indoor allergen exposure, parental allergic sensitization

EM: None

No justification givenANOVA
Njå et al., Norway [50]Follow-up study of 6- to 16-year-olds; n = 4585Fruits, vegetables, vitamin supplementation, cod liver oil supplementation, breastfeedingQuestionnaire-based asthma and wheeze; SPT

C: age, sex, area, parental education, parental atopy, exposure to dog and cat, maternal smoking during pregnancy and at child's 0–11 months

EM: None

No justification givenMultiple regression
Okoko et al., UK [51]Cross-sectional study of 5- to 10-year-olds; n = 2640FruitsISAAC-based asthma and wheeze

C: sex, age, paracetamol ibuprofen exposure, vitamin use, living in the farm, mold or mildew in bedroom, in living areas, and in hallways, financial source for home repairs, exposure to passive smoking, ethnicity, birth weight, breastfeeding, number of parents living with child, siblings, parental education,

EM: None

No justification givenMultiple regression using propensity scores of the selected covariates
Omata et al., Japan [52]

Case–control study of 3- to 15-year-olds

Cases: n = 27 Controls: n = 25

Selenium concentration in urineReported atopic dermatitis

C: None

EM: None

NANA
Pastorino et al., Brazil [53]

Case–control study of 13- to 14-year-olds;

Cases: n = 141

Controls: n = 387

Fresh fruits and cooked vegetablesISAAC-based asthma

C: Pertussis disease, siblings, cesarean section, rhinoconjunctivitis, eczema, allergic mother, SPT positivity, rhinitis, prematurity

EM: None

Empirical: based on statistical test (P < 0.20) in relation to the outcomeMultiple regression
Pesonen et al., Finland [54]Cohort study of 20-year-olds; n = 200Plasma retinol concentrationAtopic dermatitis, allergic rhinoconjunctivitis, wheeze, SPT, IgE

C: sex, breastfeeding, family allergic history, smoking in household during year 1, personal smoking at 20 years, maternal education, maternal age, sibship, day care during year 1

EM: None

No justification givenMultiple regression
Powell et al., UK [55]

Case–control study of 18-mo- to 16-yr-olds

Cases: 37

Controls: 35

Plasma tocopherol, retinol, vitamin C, and seleniumClinical-documented asthma

C: age, sex

EM: None

No justification givenSimple regression analysis
Romieu et al., the United States [56]Cross-sectional study of 2- to 16-year-olds; n = 7904Vitamin C intakeQuestionnaire-based asthma and wheeze

C: BMI, sex, race, parental asthma, parental hay fever, passive smoking, poverty level, presence of atopy, TV watching

EM: age

No justification given

C: Multiple regression

EM: Stratification

Rubin et al., the United States [57]Cross-sectional study of 4- to 16-year-olds; n = 7505Serum vitamin E, β-carotene, vitamin C, and seleniumQuestionnaire-based asthma

C: BMI, education level of household head, household crowding, urban residence, passive and active smoke exposure, parental asthma/hay fever, child history of hay fever, avoidance of pets because of allergy

EM: cigarette smoke exposure

No justification given

C: Multiple regression

EM: Interaction

Scwartz & Weiss, the United States [58]Cross-sectional study of ≥30-year-olds; n = 9074Serum vitamin C, dietary vitamin D, fish intake, dietary sodium/potassium, serum zinc/copperQuestionnaire-based bronchitis and wheeze

C: cigarette smoking, age, race, triceps skin fold, sex, family income, residence inside city center, head of household education, region

EM: age, race, sex, smoking

Conceptual: based on the previous literature.

C: Multiple regression

EM: Interaction

Shaheen et al., UK [59]Cohort study of 7.5-year-olds; 14062Dietary patternsQuestionnaire-based eczema, wheeze, hay fever, asthma; lung functions, and SPT

C: maternal smoking, infections, maternal antibiotic and paracetamol use, maternal education, housing tenure, financial difficulties, pre-pregnancy BMI, ethnicity, parity, maternal asthma, eczema, and rhinoconjunctivitis, maternal migraine, sex, gestation at birth, breastfeeding during first 6 months, day care by 8 months, multiple pregnancy, pets in infancy, damp/condensation/mold, child exposure to smoking, season of birth, season of FFQ, birth weight, head circumference, birth length, siblings, and BMI

EM: None

No justification givenMultiple regression using propensity scores of the selected covariates
Shaheen et al., UK [60]Cohort study of 30- to 42-month-olds; n = 2973Umbilical cord trace elements and mineralsQuestionnaire-based wheeze and eczema

C: sex, maternal age, parity, gestation, maternal smoking during pregnancy, maternal education, maternal housing tenure, maternal ethnicity, birth weight, maternal atopy, head circumference, crown-heel length, maternal BMI, breastfeeding during 6 months, day care by 6 months, analgesic use, and infections in late pregnancy

EM: None

No justification givenMultiple regression
Shaw et al., New Zealand [61]

Nested case–control study of 8- to 13-year-olds

Cases: 26

Controls: 61

Serum seleniumQuestionnaire-based wheeze

C: None

EM: None

NANA
Tabak et al., The Netherlands [62]Cross-sectional study of 7- to 13-year-olds; n = 598Fruits, vegetables, dairy products, whole grain products, fishISAAC-based asthma and wheeze; BHR and SPT

C: maternal education, foreign descent, age, sex, siblings, asthma in siblings, asthma in parents, maternal smoking, breastfeeding, passive smoking, pets at home, BMI

EM: None

No justification givenMultiple regression
Tahan & Karakukcu, Turkey [63]

Case–control study of 1- to 3-year-olds;

Cases: n = 34

Controls: n = 14

Zinc (in hair) deficiencyQuestionnaire-based wheeze

C: None

EM: None

NANA
Tamay et al., Turkey [64]Cross-sectional study of 6- to 12-year-olds; n = 2387FoodsISAAC-based allergic rhinitis

C: sex, family atopic history, food allergy, eczema, respiratory tract infection, frequent sinusitis, tonsillectomy, adenoidectomy, antibiotics and paracetamol use in year 1, maternal smoking, cat and dog at home at year 1, dampness at home, heating system, born in Istanbul, time lived in Istanbul, trucks passed on weekdays, perinatal redness

EM: None

No justification givenMultiple regression
Tsai et al., Taiwan [65]Cross-sectional study of 11- to 12-year-olds; n = 2290FoodsQuestionnaire-based asthma

C: residential districts, sex, doctor-diagnosed allergies

EM: None

No justification givenMultiple regression
Willers et al., The Netherlands [66]Cohort study of 8-year-olds; n = 2832FoodsISAAC-based asthma and wheeze

C: sex, maternal education, parental allergy, maternal smoking during pregnancy, smoking in the home at 8 years, breastfeeding, older siblings, birth weight, maternal overweight at yr 1, maternal supplement use during pregnancy, region, study arm

EM: None

No justification givenMultiple regression
Willers et al., UK [67]Cohort study of 5-year-olds; n = 1924FoodsISAAC-based asthma, wheeze, eczema, hay fever

C: maternal age, paternal social class, maternal age at leaving full-time education, maternal smoking during pregnancy, maternal asthma and atopy, birth weight, sex, siblings, breastfeeding, smoking in the child's home at 5 years

EM: None

No justification givenMultiple regression
Wong et al., China [68]Cross-sectional study of 10-year-olds; n = 10902Fruits and vegetablesQuestionnaire-based asthma

C: propensity scores of several unspecified covariates and sex

EM: None

No justification givenMultiple regression using propensity scores of the selected covariates
Table 2. Summary statistics on confounding and effect modification, extracted from 1
Characteristics of studiesStudy designs
All N = 62Cohort n = 20Case–control n = 16Cross-sectional n = 25Ecologic n = 1
  1. anova, analysis of variance.

Studies without adjustment for either confounders or effect modifiers90900
Studies with adjustment for either confounders or effect modifiers53207251
Studies with adjustment for only confounders40166171
Studies with adjustment for both confounders and effect modifiers134180
Studies without justification for the inclusion of confounders or effect modifiers39155190
Studies with justification for the inclusion of confounders or effect modifiers145261
Studies with conceptual justification72131
Studies with empirical justification103230
Studies with both conceptual and empirical justifications31110
Approaches to treatment of confounders in the studies
Studies with regression modeling50204251
Studies with matching20200
Studies using anova10100
Approaches to treatment of effect modifiers in the studies
Studies using stratification104150
Studies using interaction82150
Studies using both stratification and interaction52120

Time-line of studies

The earliest of these studies were conducted in 1990 [one cross-sectional study [24] and one case–control study [59]]. Subsequent studies were conducted in 1994 [two case–control studies [56, 62]]. One cross-sectional study was carried out in 1997 [19], and after that, the remaining studies were conducted between 2000 and 2010, with the first cohort studies emerging in 2004 [39, 46, 61](Table 1).

Selection of confounders and effect modifiers

Of the 62 studies reviewed, 20 utilized cohort design; 25 employed cross-sectional design; 16 were case–control studies; and one study was of ecologic design (Table 2). Nine of the studies [26, 36, 37, 40, 41, 49, 53, 62, 64], which were all case–control studies, did not take into account confounding or effect modification in the analysis. Of the studies that considered either confounding or effect modification (n = 53), 40 considered only confounding [9, 10, 15-25, 27, 30, 31, 34, 35, 38, 39, 42-45, 47, 48, 50-52, 54-56, 58, 61, 63, 65-69], while 13 considered both confounding and effect modification [8, 11-14, 28, 29, 32, 33, 46, 57-59]. Thirty-nine of these studies did not have any justification for the selection of confounders or effect modifiers included in the studies: 15 (38%), five (13%), and 19 (49%) of these studies were cohort, case–control, and cross-sectional studies, respectively (Table 2). Of the 14 studies (five cohort, two case–control, six cross-sectional, and one ecological) that justified the selection of the confounders or effect modifiers, seven employed a conceptual approach (i.e., based on knowledge of the subject matter or previous evidence in the literature), 10 employed an empirical approach (usually based on statistical tests), while only three employed both conceptual and empirical approaches (Table 2).

Although the number of confounders considered in the studies varied considerably, the most common were age, sex, familial or parental history of atopic disease, parental or household socioeconomic status (education, occupation, income), smoking (maternal or household), siblings (or parity or birth order). Some studies also considered birth weight, length, gestational age, use of antibiotics, and paracetamol. The most common effect modifiers considered included sex, smoking, history of atopy, place of residence, and race (data not shown).

None of the studies employed restriction for controlling confounding in the design stage of the study. In the handling of confounding in the statistical analysis, regression modeling was the most common approach utilized, this being employed by all the cohort [12, 14, 18, 20-22, 25, 30, 39, 43, 44, 46-48, 55, 60, 61, 67] and cross-sectional [10, 11, 15-19, 27, 29, 31-33, 38, 42, 45, 51, 52, 57-59, 63, 65, 66, 69] studies and four of the case–control studies [13, 35, 54, 56]; two [9, 24] case–control studies employed matching and one [50] employed anova. Of the studies that considered effect modification (n = 13), 10 employed stratification; eight tested for interaction, while five used both stratification and interaction (Table 2).

Summary of the comprehensive set of confounders and effect modifiers that should be considered

Based primarily on a careful examination of the factors that were considered in the appraised studies and evidence from other studies [62, 74-88], we propose in Table 3 a comprehensive set of confounders and effect modifiers that should be considered in future epidemiologic studies investigating the role of diet in the development of childhood asthma and atopic disease. These factors can be grouped mainly into maternal characteristics; birth measurements; socioeconomic characteristics; and environmental exposures. These potential confounders have been grouped into primary and secondary confounders. For the primary confounders, we suggest that these are factors that must always be considered based on conceptual justification, while the secondary confounders are factors that need to be confirmed by performing appropriate statistical tests.

Table 3. Summary of comprehensive set of confounders and effect modifiers that need to be considered in future epidemiologic studies
Confounding/effect modificationVariables
  1. a

    Always to be considered based on conceptual justification.

  2. b

    To be considered based on empirical evidence (i.e., by appropriate statistical tests).

Confounders
Primary confoundersa

Maternal and child characteristics: maternal age, child's age, child's sex, family history of atopic disease, parity, birth weight, gestational age at birth, and mode of delivery

Socioeconomic status: preferably, at least two parameters relating to education, income, occupation, or a measure of deprivation or social class

Environmental exposures: use of antibiotics and paracetamol during pregnancy and early life, pets at home, smoking (maternal smoking, household, or environmental smoking exposure), home environment (dampness, mold, chemicals), exposure to day care during infancy

Dietary factors: breastfeeding, and other relevant dietary factors depending on the research question

Secondary confoundersbMaternal BMI, living on a farm, race/ethnicity, place (or region) of birth, season of birth, and season of ascertainment of dietary factors, physical activity
Effect modifiersFamily history of atopic disease, season of birth or season of ascertainment of dietary factors, sex, race/ethnicity, place of residence, breastfeeding

Discussion

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author contributions
  10. Funding
  11. Ethical approval
  12. Coflict of interest
  13. References

The evidence from this comprehensive review of the international evidence indicates a general inadequacy in the handling of confounding and effect modification in observational studies investigating the potential effects of diet during pregnancy, infancy, and childhood on the development of childhood atopic disease and asthma. A majority of the case–control studies did not adjust for confounding or consider effect modification, while all the cohort, cross-sectional, and ecologic studies did. Of the studies that considered confounding or effect modification, most gave no justification for including the selected variables; this was the case irrespective of the study design employed. Empirical justification (usually through statistical tests), compared to conceptual justification (based on knowledge of the subject matter or evidence in previous literature), was the major criteria used in selecting the confounders and effect modifiers among the studies that provided justifications for the variables included. All cohort, cross-sectional, and ecologic studies applied regression modeling for confounder adjustment, while the case–control studies used either matching, regression modeling, or anova. Effect modification was handled either by stratifying by the effect modifiers or by including interaction terms between the effect modifiers and the exposures of interest.

Confounding constantly poses great challenges to observational epidemiologic studies and, if not efficiently handled, invariably weakens the validity of conclusions [2-5]. Effect modification, on the other hand, although not a bias aimed to be eliminated in a study [72], helps to identify whether the effect of an exposure on an outcome of interest varies across subgroups of the study population [70-72] and, in clinical practice, guides in more specific targeting of intervention to the subgroup of the population that needs it most [89]. Consequently, efficient consideration of confounding and effect modification is essential in order to reach a valid and reliable interpretation of findings from observational studies [2-5]. As highlighted in our recent systematic review, available evidence on the role of diet during pregnancy, infancy, and childhood in the development of childhood atopic disease and asthma was largely inconclusive for several dietary factors, and this was largely attributed to the methodological weaknesses of the studies [7].

The present critical appraisal of these studies shows major inadequacies concerning the handling of confounding and effect modification, particularly as it relates to the process of selection and the criteria utilized for that purpose. Poor treatment and report of confounding and effect modification has also been noted in other research contexts [5, 90, 91]. The lack of consideration of confounding in the majority of the case–control studies considered in the present paper raises serious concerns about the conclusions of these studies, because the observed effect cannot be independently attributed to the dietary factors examined without taking into account the influence of other potential factors that usually confound the diet–disease relationship. It is also surprising that a majority of studies that considered confounding or effect modification gave no inclusion criteria for the selected variables. It is possible that some of these studies, a priori, had no criteria for selecting the variables, or might have used conceptual and/or empirical criteria, but failed to report them when published. While the inclusion of confounders and effect modifiers in a study must be justified, the criteria used for their selection must also be reported in the paper. A detailed recommendation for indicating such information in a research paper has been described elsewhere [6].

Efficient treatment of confounding and effect modification, beginning from selection to treatment in statistical analyses, is also addressed in most epidemiologic texts [72]. Primarily, selection of the variables should be based on knowledge of the causal mechanism between diet and the development of allergy and asthma (conceptual justification) and secondarily on the evidence available from the data, which is usually assessed using appropriate statistical tests (based on P-values), percentage change in the main effect under study, stepwise regression models, or the more advanced techniques such as propensity scores or the shrinkage method (empirical justification) [72, 92]. Unfortunately, of the studies that gave justification for inclusion, the majority based their criteria only on empirical evidence, ignoring the primary conceptual basis. The adjustment for confounding was suitably made in the cohort and cross-sectional studies, and in some case–control studies, by employing multiple regression modeling. However, regardless of the sophistication of the method used to adjust for confounding, findings may still remain invalid if appropriate steps have not been employed in confounder selection [72]. As a result, the issue of residual confounding may remain an important concern [72]. Although matching is one of the methods used to address the issue of confounding, its limitations, particularly for case–control studies, have been extensively addressed in the literature and authors should consider these recommendations when considering matching [72, 93]. The selection of effect modifiers is mostly guided by prior knowledge of the effect under study [92]. Handling effect modification by stratification is always determined by the adequacy of the sample size of a study [89]. However, regardless of the sample size of a study, including interaction terms between the potential effect modifier(s) and the exposure of interest and performing appropriate statistical tests may be a good starting point to investigate the possible presence of effect modification [73]. The statistical significance of such tests may give a clue as to whether the effect measure is modified by the potential effect modifier(s) [73]. However, when such subgroup analyses are performed, appropriate steps should be taken to address the issue of multiple tests of association so as to minimize the risk of generating false-positive findings.

We observed that the confounders and effect modifiers considered differed between studies. This lack of uniformity in the identification and inclusion of confounders or effect modifiers in studies of diet, allergies, and asthma is likely to have contributed to the heterogeneity of reported associations. Although we acknowledge that some confounding or effect-modifying variables may be specific to a study setting, it is important that, conceptually, a comprehensive set of variables be considered by studies investigating the effect of early life diet on the occurrence of atopic disease and asthma should be proposed to guide authors when planning their study and analysis. For this purpose, based on careful consideration of the reviewed studies and other related evidence [1, 62, 74-88], we have proposed several factors that should be taken into account, in the areas of maternal characteristics, birth measurements, socioeconomic characteristics, and environmental exposures. The confounders are grouped into primary (factors that are biologically relevant in the risk of allergies or that have been consistently established in previous studies) and secondary (factors with probable or insufficient evidence in previous studies) factors, indicating that the primary confounders must always be considered, while the secondary factors should be considered based on appropriate statistical tests. For effect modification, depending on the adequacy of the sample size of a study, we propose that the following variables should be considered as potential effect modifiers: family history of atopic disease, season of birth or season of ascertainment of dietary factors, race/ethnicity, place of residence, breastfeeding. We recommend employing tests for interaction, at least in the first place, before the consideration of stratification by the effect modifiers. We acknowledge that while these lists may not be exhaustive, the proposed factors, at present, constitute a core comprehensive set of important potential confounders and effect modifiers in this area of research.

A major challenge in nutritional epidemiology is the complex inter-relationships and interactions between dietary factors both within foods and dietary habits, which ultimately may confound the observed effect of interest [94]. Consequently, including several dietary factors in the same statistical model could result in multicollinearity and hence distortion of the overall model performance. For nutrients, the observed effect may reflect interactions with other nutrients, as well as with other nonquantified nutrients. Although there are no straightforward approaches to dealing with this situation, there is a need to apply more advance statistical techniques that can account for multicollinearity. If the analysis is based on the food level, suggestions that have been offered include (i) using food groups as the unit of dietary intake instead of using single foods and (ii) employing dietary pattern analysis, such as factor or cluster analysis [94]. The situation may be less clear with nutrients considering the potential interactions between nutrients; however, appropriate statistical methods should be employed to explore such interactions. Our recent systematic review showed that the key dietary factors related to the risk of allergies in childhood include vitamins A, D, E, zinc, fruits, and vegetables [7]. With appropriate statistical methods, we suggest that these dietary factors should be considered for adjustment in subsequent dietary and nondietary studies. Furthermore, with appropriate statistical techniques, we also suggest that studies investigating the role of maternal diet should adjust for the dietary intake of the child and vice versa.

This is, as far as we are aware, the first study to address the issues of confounding and effect modification in studies investigating the possible effect of diet on the development of childhood asthma and atopic disease. We envisage that this groundwork will stimulate active discussions on the optimal strategies to consistently reduce the potential for bias in this area of research that has been mainly investigated using observational epidemiologic study designs. These observations and suggestions presented here could be the basis for further discussions and consensus on a core comprehensive set of potential confounders and effect modifiers that need to be considered when planning future epidemiologic studies. However, we note that considering confounding/effect modification–related issues will not address all of the challenges associated with observational studies, and a number of other factors, such as reverse causation, selection bias, information bias (regarding both the ascertainment of dietary intake and measurement of allergic outcomes), and sample size, still need to be considered. In addition to inadequate consideration of confounding, it is possible that differences in addressing this spectrum of factors might also be responsible for the inconsistent findings noted in previous studies. The studies included in this paper were those selected based on a set of rigorous inclusion criteria. Consequently, it is possible that the inadequacies regarding the handling of confounding and effect modification as highlighted in this paper might have been underestimated, considering that other studies with poorer quality were excluded. Regardless, we believe that the framework outlined in this paper is adequately applicable to all studies in this area of research.

Conclusions

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author contributions
  10. Funding
  11. Ethical approval
  12. Coflict of interest
  13. References

This systematic review–based analysis provides evidence that the issues of confounding and effect modification have on the whole been inadequately handled in observational epidemiologic studies investigating the role of diet in the development of childhood asthma and allergic disease. The selection of confounders and effect modifiers should be primarily based on conceptual justification and secondarily on empirical evidence. For effective handling of confounding and effect modification with consequent improvement in the methodological quality of studies in this area of research, there is a need to agree on a comprehensive set of variables that should be taken into account when planning future epidemiologic studies.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author contributions
  10. Funding
  11. Ethical approval
  12. Coflict of interest
  13. References

We would like to thank an international panel of experts for providing information on unpublished/ongoing studies. Also, our thanks go to Anna Wierzoch for administrative support. We are also grateful to the three anonymous reviewers for their constructive criticisms on an earlier draft of this manuscript.

Author contributions

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author contributions
  10. Funding
  11. Ethical approval
  12. Coflict of interest
  13. References

AS conceived the idea for this work and oversaw all aspects of the data analysis, interpretation, and writing-up. UN and BIN undertook the analysis and drafted the report. GD contributed to the interpretation of data and the writing of the manuscript.

References

  1. Top of page
  2. Abstract
  3. Background
  4. Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author contributions
  10. Funding
  11. Ethical approval
  12. Coflict of interest
  13. References