Skip Navigation

Strengthen the Evidence for Maternal and Child Health Programs

Sign up for MCHalert eNewsletter

Evidence Tools
MCHbest. Postpartum Mental Health Screening.

MCHbest Logo

Strategy. Machine Learning

Approach. Collaborate with Universities/Schools of public health to develop a machine learning algorithm to identify women that screen positive during the first year following a live birthing using electronic health records.

Return to main MCHbest page >>

Overview. Interventions involving the use of machine learning to predict and identify women screening positive for postpartum depression or anxiety have the capability of improving clinical practice and reducing socioeconomic barriers associated with accessing screening.[1] Collaborations with University hospitals and/or Schools of Public Health may offer opportunities to utilize electronic health data to strengthen the development of machine learning algorithms.[2,3] Using predictive models also has the potential to identify positive screens earlier on, potentially improving the timeliness of care.[4]

Evidence. Emerging Evidence. Strategies with this rating typically trend positive and have good potential to work. They often have a growing body of recent, but limited research that documents effects. However, further study is needed to confirm effects, determine which types of health behaviors and conditions these interventions address, and gauge effectiveness across different population groups. (Clarifying Note: The WWFH database calls this "mixed evidence").

Access the peer-reviewed evidence through the MCH Digital Library or related evidence source. (Read more about understanding evidence ratings).

Source. Peer-Reviewed Literature

Outcome Components. This strategy has shown to have impact on the following outcomes (Read more about these categories):

  • Mental Health. This strategy promotes emotional, psychological, and social well-being of individuals and communities.
  • Utilization. This strategy improves the extent to which individuals and communities use available healthcare services.
  • Access to/Receipt of Care. This strategy increases the ability for individuals to obtain healthcare services when needed, including preventive, diagnostic, and treatment services.

Detailed Outcomes. For specific outcomes related to each study supporting this strategy, access the peer-reviewed evidence and read the Intervention Results for each study.

Intervention Type. Collaboration (Read more about intervention types and levels as defined by the Public Health Intervention Wheel).

Intervention Level. Community-Focused

Examples from the Field. There are currently no ESMs that use this strategy. As Title V agencies begin to incorporate this strategy into ESMs, examples will be available here. Until then, you can search for ESMs that have similar intervention components in the ESM database.

Sample ESMs. Here are sample ESMs to use as models for your own measures using the Results-Based Accountability framework (for suggestions on how to develop programs to support this strategy, see The Role of Title V in Adapting Strategies).

Quadrant 1:
Measuring Quantity of Effort ("What/how much did we do?")

PROCESS MEASURES:

  • Number of collaborations and partnerships led by Title V established between healthcare systems, universities, and schools of public health to develop and implement machine learning algorithms for predicting postpartum depression and anxiety. (Measures the multi-sector engagement and expertise in the strategy)
  • Number of machine learning models and algorithms developed and tested to predict and identify individuals at risk for screening positive for postpartum depression or anxiety during the first year after a live birth. (Assesses the technical development and iteration of the strategy)

Quadrant 2:
Measuring Quality of Effort ("How well did we do it?")

PROCESS MEASURES:

  • Percent of collaborations and partnerships led by Title V established between healthcare systems, universities, and schools of public health to develop and implement machine learning algorithms for predicting postpartum depression and anxiety. (Measures the multi-sector engagement and expertise in the strategy)
  • Percent of machine learning models and algorithms developed and tested to predict and identify individuals at risk for screening positive for postpartum depression or anxiety during the first year after a live birth. (Assesses the technical development and iteration of the strategy)

Quadrant 3:
Measuring Quantity of Effect ("Is anyone better off?")

PROCESS MEASURES:

  • Number of healthcare providers and staff who are trained by Title V and are proficient in interpreting and acting upon the risk predictions and screening recommendations generated by the machine learning algorithms. (Measures the workforce readiness and clinical integration of the strategy)
  • Number of machine learning algorithm development and validation process that adhere to ethical principles, data privacy and security standards, and regulatory guidelines for healthcare applications. (Assesses the responsible and trustworthy use of artificial intelligence in the strategy)

OUTCOME MEASURES:

  • Number of healthcare systems and payers that demonstrate improved maternal mental health outcomes and cost-effectiveness through the scaled implementation of machine learning-based screening and tailoring interventions. (Measures the healthcare value and return on investment of the strategy)
  • Number of communities and populations that experience significant and sustained improvements in postpartum mental health and well-being through the responsive deployment of machine learning algorithms for early identification and support. (Measures the population health and resilience impact of the strategy)

Quadrant 4:
Measuring Quality of Effect ("How are they better off?")

PROCESS MEASURES:

  • Percent of machine learning algorithm development and implementation process that actively involve and support groups of postpartum individuals, family members, and community partners as co-designers, advisors, and evaluators. (Measures the depth of patient and public engagement in the strategy)
  • Percent of machine learning algorithms and models that incorporate patient-reported data to enhance the contextual relevance of the screening and risk prediction outputs. (Assesses the integration of health outcomes and patient-centeredness principles in the technical design of the strategy)

OUTCOME MEASURES:

  • Percent of healthcare systems and payers that demonstrate improved maternal mental health outcomes and cost-effectiveness through the scaled implementation of machine learning-based screening and tailoring interventions. (Measures the healthcare value and return on investment of the strategy)
  • Percent of communities and populations that experience significant and sustained improvements in postpartum mental health and well-being through the responsive deployment of machine learning algorithms for early identification and support. (Measures the population health and resilience impact of the strategy)

Note. When looking at your ESMs, SPMs, or other strategies:

  1. Move from measuring quantity to quality.
  2. Move from measuring effort to effect.
  3. Quadrant 1 strategies should be used sparingly, when no other data exists.
  4. The most effective measurement combines strategies in all levels, with most in Quadrants 2 and 4.

Learn More. Read how to create stronger ESMs and how to measure ESM impact more meaningfully through Results-Based Accountability.

References

[1] Preis, H., Djurić, P. M., Ajirak, M., Chen, T., Mane, V., Garry, D. J., ... & Lobel, M. (2022). Applying machine learning methods to psychosocial screening data to improve identification of prenatal depression: Implications for clinical practice and research. Archives of Women's Mental Health, 25(5), 965-973.
[2] Hahn, L., Eickhoff, S. B., Habel, U., Stickeler, E., Schnakenberg, P., Goecke, T. W., ... & Chechko, N. (2021). Early identification of postpartum depression using demographic, clinical, and digital phenotyping. Translational Psychiatry, 11(1), 121.
[3] Andersson, S., Bathula, D. R., Iliadis, S. I., Walter, M., & Skalkidou, A. (2021). Predicting women with depressive symptoms postpartum with machine learning methods. Scientific reports, 11(1), 7877.
[4] Hochman, E., Feldman, B., Weizman, A., Krivoy, A., Gur, S., Barzilay, E., ... & Lawrence, G. (2021). Development and validation of a machine learning‐based postpartum depression prediction model: A nationwide cohort study. Depression and anxiety, 38(4), 400-411.

This project is supported by the Health Resources and Services Administration (HRSA) of the U.S. Department of Health and Human Services (HHS) under grant number U02MC31613, MCH Advanced Education Policy, $3.5 M. This information or content and conclusions are those of the author and should not be construed as the official position or policy of, nor should any endorsements be inferred by HRSA, HHS or the U.S. Government.