Regardless of how information is gathered, the scale is immense, as is the problem of data accuracy.
Though contact tracing may seem like a relatively new concept to the general public, the government at all levels has used it for decades. One of the larger applications of contact tracing is to ensure air travel safety by matching traveler data against known smugglers, human traffickers, terrorists or disease carriers by constructing flight histories. Health officials, too, have applied contact tracing in the past to stave off a variety of communicable diseases by painstakingly identifying the points of transmission and slowing community spread.
With the global spread of COVID-19 we see how important and needed contact tracing is. The pandemic has done far more than bring contact tracing to the forefront—it has prompted the urgent need for national contact tracing on an immense scale. The state of New York, for example, is hiring as many as 17,000 contact tracers to help in the fight against the coronavirus pandemic. Across all 50 states, a mix of information about location, travel, symptoms and health conditions must be gathered from millions of Americans and subsequently logged and analyzed.
A Data Problem of Immense Scale
Today, most contact tracing is being conducted manually, with government workers in the field, spreadsheets, people in call centers, employees working with incoming data and public health experts. These efforts are likely to be supplemented digitally to some extent, as Google and Apple enable the use of Bluetooth technology to help governments and health agencies reduce the spread of the virus.
Regardless of how information is gathered, the scale is immense, as is the problem of data accuracy. Case details such as location, health information, symptoms and travel history will be logged into various separate systems, such as hospital records and state databases. Workers in the field using spreadsheets or even clipboards may record information one way, and that same information may be entered differently into a state database.
The Need for Automation
Adding to the challenge, government workers on the ground may also have difficulty with unfamiliar names or languages. The result: dirty data that can lead to inaccuracies, delays in the ability to act on data and cost overruns due to extensive manual efforts.
The pandemic and the dire need for contact tracing have brought the need for machine learning to tackle the problem into clear relief. The only way to implement national contact tracing successfully is to implement a solution where machines do most of the heavy lifting and humans enter the equation at the end to improve accuracy.
Traditionally, organizations have used top-down, rules-based methods for handling massive data volumes. A top-down approach causes data disconnects. Let’s say there is a major spike in cases in a particular neighborhood that isn’t easily explained by the data alone. In order to figure things out, a field worker may need to visit the community and ask locals why the data doesn’t make sense. The fieldworker determines that “Joe” is a new college student in the area who doesn’t yet have an address. Joe has been crashing on friend’s couches around campus and spreading the virus.
The issue here is that the people in the field may not know about the data in context, causing delays, false reporting, and other obstacles to sound analytics. What’s needed is insight from those most familiar with the situation on the ground to clarify what’s going on.
Second, rules make sense to most of us. If X, then do Y, else if A, then do B. Simple, right? And this is workable when dealing with small amounts of data. This won’t work with the amount of COVID data that needs to be managed, accurately and quickly. Some of the questions that need to be answered are incredibly complex, for instance: How close was the person to other individuals? Was he or she wearing a mask? How long was the encounter? The problem is exacerbated because as more becomes known about the virus, the rules keep changing.
Making Data Our Ally
What’s needed is an enterprise-grade data unification method that combines machine learning and human-guided expertise to unify existing data sources such as spreadsheets, medical records and state databases with unmatched speed, scalability and accuracy. The solution needs to be able to connect data sources across all contact tracing efforts to align relevant datasets to a unified schema, “cleaning” the unified dataset through entity deduplication and mastering, and “classifying” records within the clean, unified dataset to an expert-designed taxonomy for more robust downstream analysis.
The pandemic is far from over. Given the urgency of the situation, it’s time to take a hard look at how to tackle the immense challenge of contact tracing. If we harness data at scale, we can save lives and return to some semblance of normalcy more quickly.
Michael Gormley is the head of public sector for Tamr.