Integrating “Street Response” Data: Phase 1 Findings

This was originally published on July 10, 2023.

In May, we first introduced you to a new priority project for the i-Team — integrating datasets from all the teams engaged in Coordinated Street Response in San Francisco.

Last week, we put the finishing touches on Phase 1.

TL;DR? We think this has the makings of an incredibly impactful project.

Phase 1 goal: answer 3 questions

Every good discovery process starts with questions we want to answer.

1. Do departments want to share data?
2. What does not sharing data result in? Is it actually a problem?
3. Will we actually be able to link data across departments?

Here’s how we did it…

Our process

Phase 1 centered on interviews with our key stakeholders at DEM, HSH, DPH and the Fire Department. We interviewed both program managers and data teams at each department.

SF Coordinated Street Response Program

Initial interviews were conducted in May, with follow-ups and synthesizing occurring through June.

Here’s an example of some of the questions we asked:

1. What is your department’s or program’s role in street outreach?
2. Where does the data get stored? (ie, excel file, app, database, etc)
3. What data is stored in the dataset(s)? How large is it?
4. How are individual records identified? What type of join keys may be available (ie, first and last names, date of birth, address, SS number, medical record number, etc)?
5. How long has data been collected? What programs are covered?
6. Who inputs the data? How is the data entered/ updated (I.e. by frontline workers via an app, by analysts in a spreadsheet, etc)?
7. Are any accuracy or validation checks done on the data?
8. Is there any PII/ PHI/ CJIS data in the dataset? What regulations/ privacy restrictions apply to the data?
9. Are there any impediments you foresee in sharing data?

What we found

At a high level, we found the following over the course of our interviews:

1. Departments want to share data.

Most departments are either neutral, inclined, or extremely eager to share data. We did not meet any resistance to the idea of sharing data.
One interviewee at HSH said sharing data “would be the best day of my career,” and another interviewee at Fire said sharing data was “literally a matter of life or death.”

2. Not sharing data is hindering many departments from doing their work.

Our street outreach teams are feeling the pain of siloed data. They cannot track outcomes or operate as efficiently as they would like.
For example, HSOC knows when they drop off a client at a shelter, but they don’t know if the person actually enters the shelter, or makes it into more stable housing.

3. With the information we have, data linkage seems possible.

Sharing is already happening to some degree. For example, HSH’s ONE System links certain client data to client profiles in DPH’s EPIC System, and SCRT and SORT (Fire) share data with DPH daily as well.

This bodes exceedingly well, because connecting client profiles across departments will be perhaps our biggest challenge (see more below).

4. Data sharing can have profound positive impacts on our street outreach.

Many of our street outreach teams can make immediate use of shared data for profound positive impact in their work. For example, SFFD said that being able to link data from their street outreach to medical records would allow them to track outcomes of overdose response efforts. If they knew how many people survived with one medication vs. another, they could adjust their responses accordingly and literally save lives.

The biggest challenge: linking client profiles

Each department collects PII (Personally Identifiable Information — any information that can be used to identify an individual, either directly or indirectly) such as name and date of birth, which can be used in linking.

And each department stores their data in systems of record / data warehouses that can, with the right permissions, be accessed.

So this will be easy, right?

Probably not. Just because departments store the names and DOBs of their clients, doesn’t mean that the data is consistent across departments.

For instance…

• Is “John Doe” the same person as…
• “Jon Doe”? What about “Jhon Doe”?
• Or is “John Doe” born on 1/1/1980 the same person as…
• “John Doe” born 1/11/1980?

Well, maybe they are the same person and there was a typo. Or maybe they’re not.

No linkage system is perfect, especially given the constraints of the datasets, so we believe that matching all records with high confidence is likely a very long-term, difficult project.

However, we can start with matching some reports to some degree of confidence.

We plan on establishing thresholds for different levels of confidence in matching (ex, perhaps a “low” indicates 50% confidence, “medium” indicates a 70% confidence, and “high” indicates a 90% confidence). This linkage approach will be primarily beneficial for reporting and analytical use cases (i.e. identifying how many people are being served by multiple street outreach teams and their characteristics or engagement history).

Next steps: Phase 2 Minimum Viable Product (MVP)

The ultimate goal of integrating Street Response data is to link as many records of various street outreach teams as possible, in order to better respond to the needs of homeless individuals.

For Phase 2, we will build a smaller, MVP version of this larger linkage, keeping three main tenets in mind:

1. Build momentum, learn fast and iterate quickly
2. Design, prototype, and test on a smaller scale than the entire MVP
3. Deliver value to stakeholders

We estimate this portion of the project to take approximately 6 months, focused on:

• Determining data schema
• Developing data architecture
• Create secure data warehouse
• Collect initial datasets from participating departments
• Prototype linking code and procedures

We can’t wait to share this next step of the journey with you.

This was originally published on July 10, 2023. That blog has been deprecated and is reproduced in its entirety here.