A Threshold Effect in the Relation of Stressful Life Events and Preterm Delivery Nedra Whitehead, Ph

1
A Threshold Effect in the Relation of Stressful
Life Events and Preterm Delivery Nedra
Whitehead, Ph.D.
2
Background
3

• Biological studies
• Suggest stress may affect the timing of delivery
  through
• Premature hormonal stimulation of labor or
• Immunosupression resulting in increased risk of
  chorioamnionitis
• Epidemiological studies
• have been inconsistent
• Different measures of stress related to preterm
  delivery in different studies
• Some studies find no relationship

4
A possible reason for inconsistent results the
specified model is not correct

• Two types of models have been tested
• Discrete model (Model A in figure)
• Women are dichotomized as stressed or
  not-stressed.
• Average risk is the same among all stressed
  women.
• Logistic model (Model B in figure)
• Each additional unit of stress causes a linear
  increase in the log-odds of the outcome

5
A possible model which has not been tested

• Threshold model (Model C in figure)
• Stress does not cause poor pregnancy outcome
  until a certain level is reached.
• Above the threshold, each unit of stress causes a
  linear increase in the log odds of the outcome

6
Models of Relationship between Stress and
Pregnancy Outcome
No threshold (A)
Threshold (C)
Discrete exposure (B)
Stress Level
7
Methods

• Data were collected by the Pregnancy Risk
  Assessment Monitoring System (PRAMS)
• Method described by Ulm1 was used to estimate and
  test for a threshold effect

1Ulm K. A statistical method for assessing a
threshold in epidemiological studies. Stat Med
199110341-9.
8
PRAMS Pregnancy Risk Assessment Monitoring System

• What is PRAMS?
• Ongoing state population-based surveillance
  system
• Study population women who recently delivered a
  live-born infant
• Maternal attitudes, behaviors, and experiences
  during pregnancy and early infancy
• Core and state-specific items

9
PRAMSSampling and Data Collection Methods

• Sampling frame state birth certificate files
• High-risk women are oversampled
• States annual sample size 1600-3000
• Data collected 2-6 months after delivery
• Uses Dillmans2 Total Design Method
• Questionnaire mailed 2-3 times
• Mail non-responders interviewed by telephone

2 Dillman DA. Mail and telephone surveys the
total design method. 1st ed. New York, NY John
Wiley Sons, Inc., 1978
10
States and Response Rates
State
Years of Data
No. Respondents
Response Rate
Alabama
1992 - 1995
5,646
74.9
Alaska
1990 - 1995
10,142
73.2
Florida
1993 - 1995
6,991
78.6
Georgia
1993 - 1995
5,674
71.7
Indiana
1993 - 1994
5,092
70.9
Maine
1990 - 1995
5,955
81.1
Michigan
1993 - 1995
5,122
79.7
New York
1993 - 1995
4,014
73.3
Oklahoma
1990 - 1995
10,124
73.1
South Carolina
1993 - 1995
5,881
70.3
West Virginia
1990 - 1995
9,739
79.1
Total
1990 - 1995
74,380
75.0
11
Analytic Methods

• Estimating and testing threshold
• Fit a logistic model for each possible value of
  the threshold from 0 (minimum number of events)
  to 17 (one less than maximum number of events)
• Graph the log-likelihood values by the threshold
  level for the model

12

• Estimated threshold, ?, is the threshold value
  from the model with the maximum likelihood value
• To determine if a threshold exists, test
• Null hypothesis H0 ? 0
• Alternative hypothesis H1 ? gt 0
• Test statistic Log-likelihood statistic,
• R -2 (ln L (?0) - ln L(? ?)
• For constrained parameter, ?, Pr R 0.5 the
  probability from 0 to R of the standard normal
  distribution

13

• The null hypothesis is rejected if R gt 1.64
• 95 C.I. on ? includes all values of J which
  fulfill the condition D(J) 2(ln L(?) ! ln
  L(J)) lt P21, .95

14
Other Variables Interactions in the
Regression Model

• Interactions
• Maternal age
• Maternal race
• Marital status
• Maternal education
• SES indicators
• Unintended pregnancy
• Pregnancy history
• Parity

• Other Variables
• Maternal race
• Maternal age
• Marital status
• SES indicators
• Unintended pregnancy
• Pregnancy history
• Parity
• Tobacco use
• Alcohol use

15
Risk of Poor Pregnancy Outcome by Number
of Life Events
30
25
20
Percent
15
10
5
0
0
1
2
3
4
5
7
8
9
10
11
12
13
14
15
16
17
18
6
Number of Events
Preterm ()
16
Results - Bivariate comparison

• Risk of preterm delivery increased among women
  who experienced more life events

17
Number of Events
Number of Women
Preterm Delivery ()
0
25,280
7.74
1
15,068
8.80
2
11,999
9.22
3
8,351
9.32
4
5,183
9.93
5
3,231
10.45
6
1,998
12.63
7
1,208
12.67
8
704
12.96
9
410
12.58
10
180
15.07
11
107
14.90
12
43
13.79
13
20
14.12
14
13
17.74
15
6
18.51
16
5
19.50
17
2
28.17
18
19
28.41
18
Modeling results

• Threshold effects
• Only among singleton births
• Inconsistent by parity and time period
• Threshold of 5 for multiparous women from
  1990-1993
• Threshold of 2 for primiparous women from
  1994-1995
• Association of life events with preterm delivery
• Was significant only for the two models with a
  significant threshold effect
• Was weak (OR 1.06/event, 1.07/event) even when
  significant

19
Modeling Results, cont..

• Inconsistencies in results remained when analysis
  done by state, year of birth and maternal race

20
Preterm Delivery, Singletons
10
8
6
Difference in Ln-Likelihood
4
2
0
-2
-4
-6
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Threshold Level (Number of Life Events)
Primiparas 94-95
Multiparas 90-93
Primiparas 90-93
Multiparas 94-95
21
Preterm Delivery, Multiple Births
0.5
0
Difference in Log-Likelihood
-0.5
-1
-1.5
-2
0
1
2
3
4
5
6
7
8
9
10
Threshold Level
Multi 94-95
Multi 90-93
22
Threshold Effect between Number of Life Events
and Preterm Delivery
23
Conclusions

• Threshold model may fit the relation of stress
  and preterm delivery better than model with out
  threshold among some women
• Results do not support a biological relation
  between stress and preterm delivery
• Biological effect might vary by parity or
  plurality but is unlikely to vary by time