Table 3.

Comparison between the conventional Cox proportional hazards models and the marginal structural models for assessing the association of time-varying metabolic acidosis and severe metabolic acidosis with allograft dysfunction

Cox model (unadjusted)	HR	95% CI	P value
Metabolic acidosis (HCO3− < 22 mmol/l)	2.19	1.69–2.84	<0.001
Severe metabolic acidosis (HCO3− ≤ 18 mmol/l)	3.11	2.01–4.81	<0.001
Cox model (adjusted)
Metabolic acidosis (HCO3− < 22 mmol/l)	2.00	1.54–2.60	<0.001
Severe metabolic acidosis (HCO3− ≤ 18 mmol/l)	2.49	1.56–3.99	<0.001
Marginal structural model (unadjusted)
Metabolic acidosis (HCO3− <22 mmol/l)	1.83	1.39–2.40	<0.001
Severe metabolic acidosis (HCO3− ≤ 18 mmol/l)	2.25	1.32–3.84	<0.001
Marginal structural model (adjusted)
Metabolic acidosis (HCO3− < 22 mmol/l)	1.75	1.32–2.31	<0.001
Severe metabolic acidosis (HCO3− ≤ 18 mmol/l)	2.09	1.23–3.55	0.006

CI, confidence interval; eGFR, estimated glomerular filtration rate; HR, hazard ratio; KTx, kidney transplantation.

The Cox proportional hazards models were adjusted for baseline and time-varying risk factors for allograft dysfunction (baseline confounders: primary kidney disease, sex, recipient age at KTx, donor source, decade of transplantation, dialysis vintage, graft sequence, body mass index Z-score at KTx, total number of human leukocyte antigen mismatches and delayed graft function stratified within and beyond the first year posttransplant; time-varying confounders: allograft rejection and systolic blood pressure Z-score categorized by quartiles).

For marginal structural models the inverse probability of weights model was constructed, including the time-varying confounder eGFR. Models were then adjusted for other potential risk factors for allograft dysfunction.