Identifying Medicaid pregnant women at highest risk using latent class modeling and cluster analysis

Objectives:

Relying on depression and stress as gateways, we use Latent class analysis (LCA), a sophisticated statistical method, to identify pregnant women at highest risk for poor birth outcomes. We test whether a more clinically practical method, namely cluster analysis, risk stratifies women similarly to LCA. We then test whether women of the highest risk type are more likely to have poor birth outcomes.

Materials and Methods:

Pregnant women enrolled in enhanced prenatal services were surveyed through a risk screener now in effect statewide in Michigan Medicaid Enhanced Prenatal Services. Data include very detailed information on depression (ten questions on the Edinburgh scale - EPDS) and stress (four questions on the Perceived Stress Scale). LCA is used to stratify women into risk groups according to the pattern of their responses to the depression and stress questions. We then compare the latent subgroups by presenting descriptive statistics on risk factors and birth outcomes. We also perform a similar analysis, simply separating women into “not depressed” (EPDS score less than 9) or “depressed”, and compare the two subgroups along the same risk factors and birth outcomes.

We will also use regression analysis to properly test whether the two methods of risk stratifying pregnant women using mental health as the gateway are, in fact, predictive of differences in poor birth outcomes between the different subgroups.

Results:

One of the two latent classes revealed by LCA is the highest risk subgroup:

• Half the women (vs. only 11% in the other subgroup) have five or more risk factors.

• Over 80% (vs. only 6% in the other subgroup) are likely to be depressed

• Each of the modifiable health risks (smoking, alcohol use, and drug use) is more common.

• Low birth weight and preterm birth are more common.

The cluster analysis finds that depressed women (EPDS>=9) are the highest risk women:

• Approximately 55% of the depressed women have five or more risk factors, compared to only approximately 11% among the other women

• Depressed women are more likely to smoke, use alcohol, and use drugs

• Low birth weight and preterm birth are more common among depressed women

Conclusions:

LCA and cluster analysis confirm our hypothesis that mental health is a gateway for identifying women with multiple risks and who are more likely to have an adverse birth outcome. LCA and the easier to implement cluster analysis yield similar results. The highest risk pregnant women can be very well identified based on a simple EPDS score cutoff.