Multilevel spline models for blood pressure changes in pregnancy

1
Multilevel spline models for blood pressure
changes in pregnancy
Corrie Macdonald-Wallis
2
Overview

• Background and aims
• Linear spline models of blood pressure in
  pregnancy
• Adding maternal characteristics as covariates
• Modelling weight gain and blood pressure changes
  in pregnancy together
• Summary

3
Blood pressure in pregnancy

• Hypertensive disorders of pregnancy are
  associated with maternal and offspring adverse
  health outcomes
• Aims
• To describe patterns of blood pressure change
  across pregnancy
• To investigate determinants of blood pressure
  change
• To relate gestational weight gain to blood
  pressure changes

4
ALSPAC
14,541 women living in Avon, UK with expected
delivery dates between April 1991 and December
1992 recruited Routine antenatal BP (median 14
per woman) and weight (median 12 per woman)
measurements abstracted from obstetric
records 11,789 women with singleton or twin
pregnancies who had live term births and no
previous diagnosis of hypertension or
pre-eclampsia included in BP spline models
5
Linear spline models

• Multilevel linear spline models represent
  trajectories of
• change over time with linear slopes between knot
  points

Patterns of change can be estimated for each
individual as each individual has a residual for
the intercept and each of the slope parameters
These models have been used to describe
childhood growth trajectories (McCarthy et al,
2007 Howe et al, 2010)
6
Choice of knot points for BP models

• Knot points (indicating a change in slope) were
  chosen by
• Fitting fractional polynomial curves to the data
  and using the best-fitting curve to estimate
  approximate locations of knot points
• Comparing the fit of spline models to the
  fractional polynomial curve
• Comparing the fit of the spline model predicted
  values to the actual blood pressure measurements
  across pregnancy

7
Choice of knot points fractional polynomial
curves

• used to find approximate position of knot points
  in spline models
• suggests knot points at around 22, 30 and 36
  weeks gestation

8
Linear spline models for BP change
Mean at 8 weeks (mm Hg) Average change per week (mm Hg/week) Average change per week (mm Hg/week) Average change per week (mm Hg/week) Average change per week (mm Hg/week)
Mean at 8 weeks (mm Hg) 8-18 weeks 18-30 weeks 30-36 weeks 36 weeks
SBP 112.2 (111.9, 112.4) -0.131 (-0.16, -0.10) 0.153 (0.13, 0.17) 0.281 (0.24, 0.32) 1.131 (1.06, 1.20)
DBP 66.0 (65.8, 66.2) -0.183 (-0.21, -0.16) 0.105 (0.09, 0.12) 0.459 (0.43, 0.49) 1.279 (1.23, 1.33)
9
Determinants of BP change
SBP by maternal BMI
SBP by pregnancy type
model adjusted for maternal pre-pregnancy BMI,
age, parity, smoking status, highest educational
qualification and pregnancy type
10
Determinants of BP change
SBP by parity
SBP by smoking status
model adjusted for maternal pre-pregnancy BMI,
age, parity, smoking status, highest educational
qualification and pregnancy type
11
Spline model for gestational weight gain
-----18 weeks
-----28 weeks

• Weight gain has 2 knots at 18 and 28 weeks

12
Relationships between weight gain and blood
pressure change

• Is an increase in weight in one period of
  pregnancy related to a rise in blood pressure in
  the next?
• Multivariate multilevel spline model with knots
  at 18, 29 and 36 weeks for SBP and DBP and knots
  at 18 and 29 weeks for weight
• Relationships between rates of change in weight
  and BP in different periods of pregnancy derived
  from variance-covariance matrix of the random
  effects

13
SBP changes by baseline weight/ weight gain up to
18 weeks
Weight gain up to 18 weeks
Weight at 8 weeks
14
SBP changes by weight gain 18-29 weeks and 29
weeks
Weight gain 29 weeks
Weight gain 18-29 weeks
15
Summary

• Linear spline multilevel models describe patterns
  of non-linear change in an interpretable way
• We were able to explore associations of maternal
  characteristics with blood pressure changes in
  different periods of pregnancy
• We also modelled three response variables
  together SBP, DBP and weight to investigate
  temporal relationships between changes in weight
  and blood pressure in pregnancy

16
Future work

• Latent class growth models to group particular
  patterns of blood pressure change during
  pregnancy compare these with definitions of
  gestational hypertension and pre-eclampsia
• Relate patterns of change in BP to birth weight
  of offspring and cardiovascular risk factors
  measured during childhood in offspring

17
Acknowledgements

• PhD supervisors
• Kate Tilling, University of Bristol
• Debbie Lawlor, University of Bristol
• Also
• Abigail Fraser, University of Bristol
• Scott Nelson, University of Glasgow