Automated vs. Manual Case Investigation and Contact Tracing for Pandemic Surveillance: Evidence from a Stepped Wedge Cluster Randomized Trial

Cameron Raymond, Derek Ouyang, Alexis D’Agostino, Sarah L. Rudman, Daniel E. Ho

TL;DR

This study presents the first randomized trial comparing automated (survey-based) and manual (call-based) COVID-19 case investigation and contact tracing (CICT). While automated CICT significantly reduces data completeness, it does so without exacerbating racial or ethnic disparities, offering a potentially scalable and cost-effective alternative for public health surveillance.

What Problem Does This Paper Address?

Effective public health surveillance relies on timely and comprehensive data to understand disease dynamics and target interventions. Case investigation and contact tracing (CICT) are crucial for this, but manual, call-based methods are resource-intensive and difficult to scale, leading to incomplete data and unsustainable programs. While automated approaches like mobile apps exist, they face adoption and privacy barriers. This paper addresses the need for a scalable, cost-effective, and equitable alternative to manual CICT by evaluating an automated, survey-based approach.

Methods

The researchers conducted a stepped wedge cluster randomized trial in Santa Clara County, California. In this design, ZIP codes progressively transitioned from manual CICT to automated CICT over a period of several months. This allowed for a rigorous comparison of the two methods by observing changes in data completeness as each cluster (ZIP code) stepped down to the automated system. The study focused on eleven key data fields related to demographics and disease dynamics. They employed both intention-to-treat (ITT) and local average treatment effect (LATE) analyses to account for potential non-compliance and estimate the causal impact of the automated intervention.

Datasets

The study utilized data from 106,522 positive COVID-19 cases across 29 ZIP Codes in Santa Clara County, California, collected between December 15, 2021, and February 5, 2022. This data was sourced from the state-managed system used for recording and investigating COVID-19 cases (CalCONNECT).

Key Insights and Ideas

  • The Trade-off is Real: Automated CICT, while promising for scalability, comes at a significant cost to data completeness. This highlights a fundamental trade-off between efficiency and data richness.
  • No Exacerbation of Disparities: A critical finding is that the reduction in data completeness with automated CICT did not disproportionately affect any racial or ethnic group. This is crucial for ensuring that public health interventions remain equitable.
  • Scalability and Cost-Effectiveness: The paper strongly suggests that automated surveys are orders of magnitude more cost-effective and scalable than manual CICT, making them a viable option for resource-constrained public health departments.
  • "First Screening" Potential: Automated CICT could serve as an initial screening tool, freeing up human resources for more complex cases or other critical public health services.

Key Results

Automated CICT reduced the overall collected information by 29 percentage points (SE = 0.08, p < 0.01). Specifically, there were significant reductions in the completeness of symptom information (40pp) and employer information (32pp). However, no evidence of differences in information loss was found based on race or ethnicity.

Limitations and Future Work

The authors acknowledge several limitations:

  • The Omicron surge led to low manual CICT coverage, limiting the ability to study outcomes related to disease transmission rates.
  • While race and ethnicity data could be analyzed for differential effects due to imputation methods, similar analyses for other data fields were not possible.
  • The study did not assess existing biases in data collection methods for manual CICT, and it's possible that some subgroups are equally hesitant to share information regardless of the method.
  • The generalizability to other populations may be limited due to small sample sizes for certain racial/ethnic categories.

Future work could explore combined models of automated CICT with resource linkage, particularly for vulnerable populations. Further research is also needed to investigate informational gaps for specific subgroups and to understand existing biases in manual CICT data collection.

Summary updated

BibTeX
@inproceedings{raymond2022automated,
  author = {Cameron Raymond and Derek Ouyang and Alexis D’Agostino and Sarah L. Rudman and Daniel E. Ho},
  title = {{Automated vs. Manual Case Investigation and Contact Tracing for Pandemic Surveillance: Evidence from a Stepped Wedge Cluster Randomized Trial}},
  booktitle = {eClinicalMedicine},
  year = {2022},
  url = {https://openreview.net/forum?id=oPGU8ZdExH},
}
Figure