Aim
We endeavor to develop a multiple logistic regression prediction model to estimate risk of late-pregnancy stillbirth from 35 weeks using a national dataset of all births in Australia to ultimately inform decision-making around timing of birth.
Study design
This is a protocol for a retrospective cohort study using the total population of singleton term gestation births in Australia (1998–2015) included in the National Perinatal Data Collection (NPDC)(10). The dataset includes 7,200 stillbirths among 4.9 million births at an estimated rate of 1.47 stillbirths per 1000 births (10). Multiple pregnancies and congenital abnormalities will be excluded. A congenital abnormality is defined as a stillbirth classified as code 0100 “Congenital Abnormality” using the Perinatal Society of Australia and New Zealand (PSANZ) Perinatal Death Classification System (19). Candidate predictors identified in a meta-analysis and through clinical consultation will be assessed for inclusion in the full model to ensure adjustment prior to reverse stepwise elimination.
Data source
All births from 35 weeks gestation in Australia (1998–2015) included in the NPDC will be included and made available via the AIHW Maternal and Perinatal Health Unit. The NPDC is a national population-based cross-sectional collection of data for all pregnancies and births established in 1991 (20). The NPDC includes all births from the 6 states and 2 territories of Australia. Perinatal data are collected for each birth in each state and territory, usually by midwives and other birth attendants (10). The data is collated by the relevant state or territory health department and a standard de-identified extract is provided to the AIHW on an annual basis to form the NPDC (10). Stillbirths in Australia are defined by the PSANZ as fetal deaths from gestational age of at least 20 weeks or birthweight of at least 400 grams, except in Victoria and Western Australia, where births are included if gestational age is at least 20 weeks or, if gestation is unknown, birthweight is at least 400 grams (10, 19).
Model development
A multiple logistic regression model will be developed where the outcome (stillbirth) is binary and the independent variables are either continuous or categorical. Reference group coding will be informed by literature and existing reporting recommendations. Descriptive statistics will be used to characterize the study cohort and illustrate censoring and survivability throughout gestation.
The predictor selection process is illustrated in Fig. 1. Previous meta-analyses have established characteristics and conditions associated with an increased risk of stillbirth that will be considered as candidate predictors (15, 21–23). Frequencies (%) will be presented for categorical variables and mean/standard deviation, median/interquartile range (IQR), minimum, and maximum will be presented for continuous (numerical) variables. If clinically appropriate and statistically justifiable, independent continuous variables will be logically grouped according to published guidelines and recommendations (10, 24).
We will first fit a multivariable logistic regression model containing all candidate predictors (Table 1). Clinical background knowledge and existing meta-analyses will further inform variable selection for the final model. In addition, reverse stepwise elimination will be explored to remove non-significant factors with P-values greater than 0.100 in line with Akaike’s Information Criterion (25). Redundant variables demonstrating multi- or collinearity will be logically excluded through clinical consultation. Single value imputation will be considered for predictors with greater than 5% missing values or will be excluded. To alleviate risk of bias in the model, no births will be excluded due to missing data.
Table 1
Original dataset characteristics of candidate predictors for all births in Australia, 1998–2015.
Candidate predictor | Reported value | Reported label |
Model of care (mother insurance status) | 1 | Public |
2 | Private |
8 | Not applicable (e.g., home birth) |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Hospital sector | 1 | Public |
2 | Private |
8 | Not applicable (e.g., home birth) |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Maternal age in years | < 19 | Less than 19 |
19–44 | Single values |
>=45 | 45 and over |
99 | Not stated |
Maternal Indigenous status | 1 | Indigenous - Aboriginal and/or Torres Strait Islander |
2 | Non-Indigenous |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Maternal country of birth | ASCSS, SACC codes | ASCSS, SACC 1st edition, SACC 2nd edition, SACC 2011 or pre-arranged groupings |
Pre-existing diabetes during pregnancy | 0 | None/not stated |
1 | Pre-existing diabetes |
Missing | Jurisdiction did not provide any data |
Gestational diabetes | 0 | None/not stated |
1 | Pre-existing diabetes |
2 | Gestational diabetes mellitus (GDM) |
Missing | Jurisdiction did not provide any data |
Chronic hypertension during pregnancy | 0 | None/not stated |
1 | Chronic hypertension |
Missing | Jurisdiction did not provide any data |
Maternal medical conditions: Essential hypertension | 0 | None/not stated |
1 | Gestational hypertension |
Missing | Jurisdiction did not provide any data |
2006 Socio-Economic Indexes for Areas (SEIFA) Index of Relative Socio-Economic Disadvantage (IRSD) | 1 | Quintile 1 (most disadvantaged) |
2 | Quintile 2 |
3 | Quintile 3 |
4 | Quintile 4 |
5 | Quintile 5 (least disadvantaged) |
9 | Not stated |
2011 Socio-Economic Indexes for Areas (SEIFA) Index of Relative Socio-Economic Disadvantage (IRSD) | 1 | Quintile 1 (most disadvantaged) |
2 | Quintile 2 |
3 | Quintile 3 |
4 | Quintile 4 |
5 | Quintile 5 (least disadvantaged) |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Remoteness Area as per the Australian Standard Geographical Classification (ASGC) | 0 | Major cities |
1 | Inner regional |
2 | Outer regional |
3 | Remote |
4 | Very remote |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Remoteness Area as per the Australian Statistical Geography Standard (ASGS) | 0 | Major cities |
1 | Inner regional |
2 | Outer regional |
3 | Remote |
4 | Very remote |
Missing | Jurisdiction did not provide any data |
Blank | Not able to be assigned, non-Australian resident and not stated |
Marital status | 1 | Never married |
2 | Widowed, divorced, separated |
3 | Married (including de facto) |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Total number of previous pregnancies resulting in a livebirth or a stillbirth | 0–4 | Single values |
>=5 | 5 and over (grouped) |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Previous caesarean sections | 0–4 | Single values |
>=5 | 5 and over (grouped) |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Caesarean section for last birth | 1 | Yes |
2 | No |
7 | Not applicable |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Previous pregnancies resulting in stillbirths | 0–1 | Single values |
>=2 | 2 and over (grouped) |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Smoking status during pregnancy | 1 | Smoked |
2 | Did not smoke |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Smoking status during first twenty weeks of pregnancy | 1 | Smoked |
2 | Did not smoke |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Smoking status after twenty weeks of pregnancy | 1 | Smoked |
2 | Did not smoke |
Missing | Jurisdiction did not provide any data |
9 | Not stated |
Pre-pregnancy Body Mass Index (BMI) | ≥ 9.0 | Continuous values rounded to the nearest 5 kg/m2 |
Missing | Jurisdiction did not provide any data |
99.9 | Not stated |
Plurality | 1–8 | Single values |
9 | Not stated |
Assisted reproduction technology flag | 0 | No |
1 | Yes |
missing | Jurisdiction did not provide any data |
9 | Not stated |
Baby's birth order | 1–8 | Single values |
9 | Not stated |
Baby's sex | 1 | Male |
2 | Female |
9 | Indeterminate and not stated |
Validation
A geographic validation approach will be considered whereby gestation-specific models will be developed using eight combinations of jurisdictionally grouped data for internal and external validation (Fig. 2). Population characteristics and performance measures will be reported for all individual models (26).
Internal validation
Internal validation will be performed using a cross-validation approach (100- or 200-fold) on all birth data for respective population groups per Fig. 1 where we anticipate greater than 100 events per predictor for populations exceeding 5000 stillbirths (27). Summary stillbirth rates will be reported to characterize fluctuating stillbirth prevalence across folds.
External validation
Final models will be externally validated using data derived from the excluded jurisdictions (28). Assessment of performance will include calibration, discrimination, positive predictive value (PPV), and negative predictive value (NPV).
Model performance
The performance of development and validation datasets will be assessed via overall performance (R2), calibration, discrimination, and clinical usefulness (PPV and NPV).
Calibration characterizes model performance in terms of agreement between predicted (expected) risk and observed risk and be visualized using a calibration plot (29). An intercept of zero and ratio of observed and expected equal to one (O/E = 1) is defined as best possible calibration (30). Confidence intervals (95%) will be prepared alongside calibration plots to visualize the degree of calibration between observed outcomes and predictions.
Discrimination is defined as the model’s ability to distinguish stillbirths and non-stillbirths and will be measured via calculation of the concordance (C) statistic and visualized with a receiver operator characteristic (AUROC) curve. An AUROC curve is used to visualize the performance of a categorical classifier and is a plot of sensitivity (true positive rate) versus 1-specificity (false positive rate) where different points on the curve correspond to different cut-off points used to designate positive identification/classification. (31). Using the AUROC curve, the performance of the predictors will be further quantified by calculating the area under the curve, or AUC. The AUC score range is 0.0–1.0, where a score of 0.5 can be equated to a ‘coin flip,’ 0.0 is perfectly inaccurate, and 1.0 is perfectly accurate (32). A non-parametric comparison of AUC will be performed using the Mann-Whitney U-statistic for individual models (26).
In addition to calibration and discrimination, PPV and NPV will be reported to characterize clinical usefulness. A decision curve analysis will be considered to visualize potential decision thresholds (33).