Methods
We obtained data from the discharge abstract database of the Canadian Institute for Health Information. This is a national database that contains information on all admissions to hospital in Canada (excluding Quebec) and includes personal details, medical history, diagnoses, and procedures associated with each hospital admission. The database has been validated and is routinely used for surveillance and research purposes. The study cohort included all hospital deliveries (n=2 193 425) in Canada (excluding the province of Quebec), that resulted in a live birth or stillbirth from April 2003 to March 2011 (hereafter referred to as 2003 to 2010).
Postpartum haemorrhage was defined using ICD-10CA (international statistical classification of diseases and related health problems (Canadian version,) codes O72.0 to O72.3, as a blood loss of ≥500 mL after vaginal delivery or ≥1000 mL after cesarean delivery, or as noted in the medical record by a care provider. Subtypes of postpartum haemorrhage were similarly identified using appropriate codes (see Supplementary Table 1 ). We defined severe postpartum haemorrhage as postpartum haemorrhage plus blood transfusion, hysterectomy, or other procedures to control bleeding. A validation study of the discharge abstract database reported a sensitivity of 90.2% and specificity of 98.2% for postpartum haemorrhage, and a sensitivity of 85.7% and specificity of 99.8% for blood transfusion.
Hypertensive disorders of pregnancy included pre-existing hypertension with or without superimposed proteinuria (O10-O11), gestational hypertension without significant proteinuria (O13), gestational hypertension with significant proteinuria (O14), eclampsia (O15), and unspecified maternal hypertension (O16). The validation study of the discharge abstract database reported a sensitivity of 87.9% and a specificity of 99.6% for gestational hypertension and a sensitivity of 83.3% and a specificity of 99.9% for pre-existing hypertension.
Acute renal failure among women admitted for childbirth was defined using standard ICD-10 codes, including those for postpartum acute renal failure (O90.4), post-procedural renal failure (N99.0), acute renal failure (N17), and unspecified kidney failure (N19). The outcome of acute renal failure was based on the healthcare provider’s documentation of this diagnosis in the woman’s medical chart. To avoid potential temporal ambiguity between the determinant and the outcome (that is, acute renal failure preceding postpartum haemorrhage), we repeated our analyses after restricting the study to cases of postpartum acute renal failure (O90.4). To enhance the stability of rate estimates we estimated temporal trends in obstetric acute renal failure by year and also in two year periods. To ensure that any temporal changes did not merely reflect changes in the occurrence or reporting of milder cases, we examined rates of maternal death, dialysis, and intensive care unit admission among women with obstetric acute renal failure.
Other determinants of acute renal failure examined in the study were maternal age and parity, multi-fetal gestation, caesarean delivery, and induction of labour. Maternal comorbidity examined included diabetes (pre-existing or gestational), gestational oedema and proteinuria without hypertension, sepsis, other puerperal infection, placenta praevia, placental abruption, unspecified antepartum haemorrhage, uterine rupture, and cardiac failure.
We first assessed temporal trends in annual rates of obstetric acute renal failure, postpartum haemorrhage, hypertensive disorders, and other risk factors using the χ test for linear trend in proportions. Among cases of obstetric acute renal failure, the frequency of these risk factors was quantified in both the first half (2003-06) and the second half (2007-10) of the study period (the test for linear trend in proportions was based on the rate for each year of study). Rates of acute renal failure were also quantified among women with and without postpartum haemorrhage, with and without hypertensive disorders, and other risk factors.
Logistic regression analyses were carried out to quantify the effects of period (2009-10 v 2003-04) on obstetric acute renal failure, while controlling for potential confounders; the contrast between 2009-10 and 2003-04 was intended to facilitate the interpretation of changes in rates over time. To determine whether the crude temporal increase was explained by changes in postpartum haemorrhage and other risk factors we compared the unadjusted and adjusted odds ratios expressing the temporal change in obstetric acute renal failure. We also modeled the temporal change in obstetric acute renal failure with calendar year of delivery (2003 to 2010) entered as a continuous variable in the logistic model.
We assessed the statistical significance of the interaction term between calendar year and gestational hypertension with proteinuria. This post hoc analysis was motivated by observed differences in temporal trends in obstetric acute renal failure among women with and women without gestational hypertension with proteinuria. We repeated analyses within strata of interest (for example, among women with and without postpartum haemorrhage, and among women with and without gestational hypertension with proteinuria). Sensitivity analyses explored whether temporal trends in obstetric acute renal failure were better explained by changes in severe postpartum haemorrhage (instead of any postpartum haemorrhage) or by modeling age as a continuous covariate using restricted cubic splines. We also carried out a post hoc assessment of temporal trends in rates of pulmonary oedema. Finally, we assessed if adjustment for other potential confounders—namely, previous caesarean delivery, chronic kidney disease, obesity, and obstetric shock—accounted for temporal changes in obstetric acute renal failure.
All analyses were conducted using the statistical software package SAS version 9.3 and Stata SE version 11.