Protocol for a nested case-control study design for omics investigations in the Environmental Determinants of Islet Autoimmunity cohort

Figures & data

Figure 1. Flow chart for inclusion of children in the nested case control study: matching of controls to cases. Note: Cases are eligible to be controls up until their date of onset of persistent IA. Fifty-three cases had three matched controls, one case had two matched controls.

Table 1. Characteristics of the NCC study participants.

	Cases	Matched controls	Eligible cohort
	(N = 54)	(N = 161)^a	(N = 1307)^b
Child sex
Female	28 (51.9%)	84 (52.2%)	633 (48.4%)
Male	26 (48.1%)	77 (47.8%)	673 (51.5%)
Missing	0 (0%)	0 (0%)	1 (0.1%)
Proband relationship
Mother	17 (31.5%)	87 (54.0%)	787 (60.2%)
Father	19 (35.2%)	45 (28.0%)	342 (26.2%)
Sibling	9 (16.7%)	21 (13.0%)	137 (10.5%)
Multiple	9 (16.7%)	8 (5.0%)	36 (2.8%)
Other	0 (0%)	0 (0%)	5 (0.4%)
Child HLA^c
DR3,4	17 (31.5%)	14 (8.7%)	140 (12.0%)
Not DR3,4	36 (66.7%)	142 (88.2%)	1023 (88.0%)
Indeterminant	0 (0%)	4 (2.5%)	4 (0.3%)
Missing	1 (1.9%)	1 (0.6%)	140 (10.7%)
Gestational age at birth (weeks)
Median (Q1–Q3)	38.5 (37.2–39.4)	38.0 (36.9–39.4)	37.9 (36.6–38.7)
Missing	0 (0%)	0 (0%)	4 (0.3%)
Mode of birth
Caesarean	33 (61.1%)	89 (55.3%)	766 (58.6%)
Vaginal	21 (38.9%)	72 (44.7%)	533 (40.8%)
Missing	0 (0%)	0 (0%)	8 (0.6%)
Birth weight (grams)
Mean (SD)	3540 (480)	3580 (535)	3500 (656)
Missing	0 (0%)	0 (0%)	8 (0.6%)
Plurality of pregnancy
Singleton	53 (98.1%)	161 (100%)	1269 (97.1%)
Twins	1 (1.9%)	0 (0%)	35 (2.7%)
Triplets	0 (0%)	0 (0%)	3 (0.2%)
Any breastfeeding^c
Yes	53 (98.1%)	156 (96.9%)	1206 (98.2%)
No	0 (0%)	5 (3.1%)	22 (1.8%)
Missing	1 (1.9%)	0 (0%)	79 (6.0%)
Apgar score at 5 min
Median (Q1–Q3)	9.00 (9.00–9.00)	9.00 (9.00–9.00)	9.00 (9.00–9.00)
Missing	0 (0%)	0 (0%)	14 (1.1%)

^aThere are 161 matched controls (Figure 1), includes duplicated information from individuals that are controls to more than one case and controls that later become cases.

^bEligible children (Figure 1).

^cFor those count variables with >5% missing, the % of Yes and No are based on those children with data (i.e. exclude missing).

Table 2. Childhood (including birth) total numbers of samples, mean (standard deviation) and median (interquartile range) number of samples per child, by sample type for cases and controls of the ENDIA nested-case control study.

Sample type	Case	Control
	(N = 54)	(N = 161)
Serum
Total samples	210	863
Mean (SD)	3.89 (2.81)	5.36 (3.15)
Median (Q1–Q3)	3.00 (2.00–5.00)	4.00 (3.00–7.00)
Plasma
Total samples	189	794
Mean (SD)	3.50 (2.56)	4.93 (3.03)
Median (Q1–Q3)	3.00 (2.00–4.00)	4.00 (3.00–6.00)
Stool
Total samples	252	1000
Mean (SD)	4.67 (2.44)	6.22 (2.84)
Median (Q1–Q3)	4.00 (3.00–6.00)	6.00 (4.00–8.00)
Urine
Total samples	202	782
Mean (SD)	3.74 (2.84)	4.86 (3.05)
Median (Q1–Q3)	3.00 (2.00–4.00)	4.00 (3.00–6.00)
Throat swab
Total samples	265	1050
Mean (SD)	4.91 (2.86)	6.53 (3.21)
Median (Q1–Q3)	4.00 (3.00–6.00)	6.00 (4.00–9.00)
Tongue swab
Total samples	273	1070
Mean (SD)	5.06 (2.82)	6.62 (3.21)
Median (Q1–Q3)	4.00 (3.25–6.00)	6.00 (4.00–9.00)
Buccal swab
Total samples	270	1060
Mean (SD)	5.00 (2.87)	6.60 (3.22)
Median (Q1–Q3)	4.00 (3.25–6.00)	6.00 (4.00–9.00)
Nasal swab
Total samples	271	1070
Mean (SD)	5.02 (2.82)	6.65 (3.21)
Median (Q1–Q3)	4.00 (3.25–5.75)	6.00 (4.00–9.00)
Skin swab
Total samples	301	1180
Mean (SD)	5.57 (2.90)	7.30 (3.24)
Median (Q1–Q3)	5.00 (4.00–6.00)	6.00 (5.00–10.0)
Breastmilk
Total samples	137	533
Mean (SD)	2.54 (1.92)	3.31 (2.04)
Median (Q1–Q3)	2.00 (1.00–4.00)	3.00 (2.00–5.00)
Total
Total samples	2370	9402
Mean (SD)	43.9 (24.9)	58.4 (27.1)
Median (Q1–Q3)	37.5 (28.3–48.8)	52.0 (38.0–73.0)

Note: Case:control ratio was 1:3 (Figure 1). Childhood samples are collected at scheduled study visits:-birth, three monthly until two years of age, then six monthly. Samples from a case included those up until the date of onset of persistent IA (inclusive). Samples from a control included those available until it’s matched case onset time.

Table 3. Pregnancy total numbers of samples, mean (standard deviation) and median (interquartile range) number of samples per mother, by sample type for cases and controls of the ENDIA nested-case control study.

Sample type	Case	Control
	(N = 39)	(N = 135)
Serum
Total samples	66	266
Mean (SD)	1.69 (0.800)	1.97 (0.872)
Median (Q1–Q3)	2.00 (1.00–2.00)	2.00 (1.00–3.00)
Plasma
Total samples	66	265
Mean (SD)	1.69 (0.800)	1.96 (0.867)
Median (Q1–Q3)	2.00 (1.00–2.00)	2.00 (1.00–3.00)
Stool
Total samples	51	235
Mean (SD)	1.31 (0.800)	1.74 (0.969)
Median (Q1–Q3)	1.00 (1.00–2.00)	2.00 (1.00–3.00)
Urine
Total samples	65	271
Mean (SD)	1.67 (0.838)	2.01 (0.806)
Median (Q1–Q3)	1.00 (1.00–2.00)	2.00 (1.00–3.00)
Throat swab
Total samples	68	272
Mean (SD)	1.74 (0.818)	2.01 (0.810)
Median (Q1–Q3)	2.00 (1.00–2.00)	2.00 (1.00–3.00)
Tongue swab
Total samples	69	275
Mean (SD)	1.77 (0.777)	2.04 (0.796)
Median (Q1–Q3)	2.00 (1.00–2.00)	2.00 (1.00–3.00)
Buccal swab
Total samples	69	275
Mean (SD)	1.77 (0.810)	2.04 (0.796)
Median (Q1–Q3)	2.00 (1.00–2.00)	2.00 (1.00–3.00)
Nasal swab
Total samples	70	275
Mean (SD)	1.79 (0.767)	2.04 (0.796)
Median (Q1–Q3)	2.00 (1.00–2.00)	2.00 (1.00–3.00)
Skin swab
Total samples	37	128
Mean (SD)	0.949 (0.223)	0.948 (0.223)
Median (Q1–Q3)	1.00 (1.00–1.00)	1.00 (1.00–1.00)
Vaginal swab
Total samples	36	118
Mean (SD)	0.923 (0.270)	0.874 (0.333)
Median (Q1–Q3)	1.00 (1.00–1.00)	1.00 (1.00–1.00)
Total
Total samples	597	2380
Mean (SD)	15.3 (5.97)	17.6 (6.54)
Median (Q1–Q3)	16.0 (10.0–18.0)	18.0 (10.0–25.0)

Note: Case:control ratio was 1:3. Mothers of siblings will appear in the table multiple times. Pregnancy samples can include samples from Trimester 1, 2 and 3. There are 26 control children with no maternal samples, 25 of these were postnatal recruits and 1 was recruited during pregnancy but had no maternal samples. There are 15 case children with no maternal samples, 13 of these were postnatal recruits and 2 were recruited during pregnancy but had no maternal samples.

Table 4. Summary of univariate and multivariate analysis approaches considered and those to be implemented for a single time and longitudinal omic data in the ENDIA study.

	Univariate^a	Multivariate^e
Single time	Unweighted conditional logistic regression^b Weighted logistic regression^b Weighted conditional logistic regression^c	Principal component analysis (PCA) Partial least square-discriminant analysis (PLS-DA)
Longitudinal	Linear or generalized linear mixed models^d IA status as response IA status as explanatory factor	N-way PLS-DA Clustering Network analysis methods

^aUnivariate methods will require false discovery rate (FDR)²⁹ adjustment with <0.1 cut-off.

^bDiscussed in Supplementary Material I.

^cThis is the primary analysis, weight calculations are in Supplementary Material J and example of R code to fit models in Supplementary Material K.

^dThe hierarchy of models fitted are discussed in Supplementary Material N.

^eMultivariate approaches will use adjusted analyte values (Table 3) and are discussed in Supplementary material O.

Table 5. Summary of potential variables included in the univariate and multivariate models.

	Explanatory variables	Design variables	Confounders
Fixed effects	Analyte^a IA status^b (factor) Time^c Children’s analysis (continuous) Mother’s analysis (factor) IA status and time interaction^c Sample ID^d (factor)	Risk-set^e (factor) Matching variables^e Age (continuous) Sex (factor)	Confounder variables or propensity score^eh
Random effects		Laboratory design (factors)^e Mother^ef (factor)
Log offset		Weights^eg

^aPrimary analysis only, may be either continuous or a factor (e.g. presence/absence).

^bSecondary analyses only.

^cLongitudinal analyses only.

^dMultivariate approach only, where adjusted sample analyte values are to be used which are determined based on a model with an explanatory variable relating to sample ID, discussed in Supplementary Material O.

^eDiscussed in Supplementary Material K.

^fInduces a correlation between NCC siblings who have the same mother.

^gCalculation of weights is shown in Supplementary Material J.

^hDiscussed in Supplementary Material L.

Lin IF, Paik MC. Matched case—control data analysis with selection bias. Biometrics. 2001;57(4):1106–1112.

 PubMed Web of Science ®Google Scholar

Data availability statement

Data sharing is not applicable to this article as no new data were created or analysed in this study. Data dictionaries and study protocols are available here: https://www.endia.org.au/for-researchers/. Data and samples can be requested by researchers who provide a methodologically sound proposal that aligns with the goals of the ENDIA Study Group as described here: https://www.endia.org.au/for-researchers/.