Serum and cerebrospinal fluid immune mediators in children with autistic disorder: a longitudinal study

Table 2 Serum results of mixed models

	Limit of detection (pg/mL)	% outside limit of detection	Within df	Test statistics from mixed model
	Limit of detection (pg/mL)	% outside limit of detection	Within df	Group (1 df)	Age (2 df)	Age² (2 df)
				F(p)	F(p)	F(p)
Cytokines
IL-1α	0.35	15	149	0.01 (.94)
IL-1RA^a	0.09	49	193	0.27 (.85)
IL-1β^a	0.03	49	193	0.23 (.85)
IL-2	0.03	29	117	0.46 (.81)
sIL-2RA^a	0.30	33	189	0.43 (.81)	7.72 (.01)	3.65 (.13)
IL-3^a	0.01	46	193	4.80 (.13)
IL-4	0.04	28	116	0.59 (.77)
IL-5	0.01	10	163	0.2 (.86)
IL-6^a	0.06	34	193	0.23 (.85)
IL-7	0.03	12	156	0.89 (.70)
IL-9	0.01	6	171	0.03 (.93)
IL-10	0.01	27	119	0.1 (.88)	3.97 (.11)	4.5 (.07)
IL-12p40	0.57	16	146	1.12 (.60)	5.81 (.04)
IL-12p70	0.05	15	147	1.27 (.60)
IL-13^a	0.01	37	193	0.62 (.77)
IL-15^a	0.17	52	193	0.40 (.81)
IL-17	0.01	5	177	0.10 (.88)
IFNα2	0.12	5	174	0.79 (.73)
IFNγ	0.04	2	186	0.02 (.93)
TNFα	na	0	190	0.02 (.93)	13.10 (<.001)
TNFβ	0.02	27	123	0.04 (.93)
TGFα	0.02	12	153	1.12 (.60)
Growth factors
EGF	2.70	7	170	8.64 (.04)
G-CSF	1.65	1	186	0.10 (.88)	3.99 (.11)
GM-CSF	1.85	2	186	0.36 (.81)
VEGF	3.36	3	185	2.26 (.43)
FGF-2	4.08	8	166	0.02 (.93)
FLT-3L^a	0.04	57	193	1.19 (.60)
sCD40L^a	1 671 700	53	193	15.0 (<.001)
Chemokines
CCL2 (MCP-1)	na	0	192	2.04 (.44)
CCL3 (MIP-1α)	0.03	8	169	1.59 (.53)
CCL4 (MIP-1β)	2.9	1	191	0.15 (.88)
CCL7 (MCP-3)	0.73	21	132	0.59 (.77)
CCL11 (EOTAXIN)	17.06	1	188	0.01 (.94)
CCL22 (MDC)	na	0	190	4.73 (.13)	7.92 (.01)
CXCL1 (GRO)	na	0	193	1.65 (.53)
CXCL8 (IL-8)	na	0	189	0.12 (.88)	1.63 (.53)	5.18 (.07)
CXCL10 (IP-10)	na	0	192	2.59 (.42)
CX3CL1 (FRACTALKINE)	2.1	26	119	0.02 (.93)

Note: df = degrees of freedom; na = not applicable. Out-of-range values were below the limit of detection for all variables except sCD40L, for which out-of-range refers to values above the limit of detection. Out-of-range values were set to missing, except where the % outside the limit of detection was ≥30%. Those variables were analyzed as categorical (out-of-range versus detectable), modeled with a binary distribution and logit link. A series of nested models were tested; sequential models for quadratic, linear, and no effect of age were tested. Where a linear or quadratic term was significant, a contrast statement was used to determine the difference in linear slope or quadratic shape between groups; in no case was this contrast statistically significant. Raw p values were used to determine the best-fitting model; FDR-adjustment was performed after this model was selected and are presented in the table; many terms did not remain significant after correction. All models included sex as a covariate and a random effect of intercept to control for observations correlated within subject
^aVariables that were analyzed as categorical (out-of-range versus detectable) and modeled with a binary distribution and logit link. The remaining variables were analyzed as Ln-transformed continuous variables with a Gaussian distribution