Thinklytics

St. David's Medical Center · Healthcare · Austin, TX · 12 weeks

Patient data quality score raised from 58 to 91 in 12 weeks

St. David’s spent $3.2M on a population health platform that couldn’t launch because patient data didn’t meet quality standards. We dug into their demographic and encounter data, found the root causes of the issues, and fixed them within 12 weeks. Our work raised their data quality score from 58 to 91, allowing the platform to go live as planned.

Challenge

The population health platform needed a data quality score of at least 80 across 10 categories to activate. St. David’s initial score was 58. Their internal data team spent six months trying to fix the issues but made no real progress. The platform was paid for but wasn’t delivering any value.

Approach

In week one, we drilled into each failing dimension to find the root cause. Six out of ten problems came from one upstream ETL job that had been dropping records unnoticed for 18 months. We corrected the ETL process, restored the missing data from source systems, and created automated dashboards to monitor quality on the other four dimensions.

Outcome

By week 12, we raised the quality score to 91, surpassing the vendor’s 80-point benchmark. Our ETL fix recovered 340,000 patient records that had been missed before. The population health platform launched three weeks after we finished. We set up continuous quality monitoring that flags new data problems within 24 hours.

We ran into a hiccup when an ETL glitch scrambled six dimension tables.

Here’s what happened: one of our ETL jobs was silently dropping records whenever the zip code was missing. This flew under the radar for a good 18 months and caused the bulk of our data quality headaches. When we finally caught it, we fixed the job, and boom, most of the problems just vanished.

How We Found 340,000 Missing Records by Hunting Down Data Bugs

Alright, here’s the deal. The EHR system was missing a whopping 340,000 records. Huge gap, right? So, we built this backfill pipeline to fix it. We ran it overnight for four nights in a row, pulling in all the missing demographic and encounter data. And the cool part? We did it without disrupting anyone’s daily work. Smooth and clean.

How we set up nonstop quality checks on autopilot

We built a Power BI dashboard to track 10 key data quality metrics every day. If any of those fall below the threshold, the system sends a notification to the data team within 24 hours. This simple setup caught data quality problems early, no more surprises slipping through the cracks like before.

Results

  • 58 to 91 Data quality score
  • $3.2M Platform investment unblocked
  • 340K Patient records recovered
  • 12 wks From assessment to go-live

We invested $3.2 million in a platform that ended up not fitting our needs. Thinklytics found the issue in the first week, had it resolved by week four, and by week fifteen, we were fully operational. Now, it’s our go-to tool for managing population health.

Chief Medical Information Officer, St. David's Medical Center

In the client's terms

The question

Where does this number come from, and can the path be trusted?

The situation

A population health platform would not activate below a data quality score of 80 across ten dimensions. The score was 58. Six months of internal effort had not moved it, and the platform was paid for and producing nothing.

What we found

Six of the ten failing dimensions traced back to one upstream ETL job that had been silently dropping records for 18 months. The quality score was not ten problems. It was one problem wearing ten masks, and nobody had looked upstream.

What it produced

The score reached 91 by week 12, against a vendor threshold of 80. The ETL fix recovered 340,000 patient records. The $3.2M platform launched three weeks after delivery. Continuous monitoring now flags new quality faults within 24 hours.

The roadmap

Week one went dimension by dimension to root cause rather than treating the score as the thing to fix. The ETL fault was corrected and the missing records were recovered from the source systems. Automated monitoring was built for the remaining four dimensions so the next silent failure surfaces in a day rather than a year and a half.

The work behind this result

Related case studies

Frequently asked questions

How quickly can a data quality score be raised to a vendor threshold?

St. David's Medical Center went from 58 to 91 in twelve weeks against a threshold of 80, after six months of internal effort had not moved it. The speed came from root-causing rather than remediating dimension by dimension.

What causes a low data quality score across many dimensions?

Often one upstream fault rather than many. At St. David's, six of the ten failing dimensions traced back to a single ETL job that had been silently dropping records whenever the zip code was missing, for 18 months. Fixing that job cleared six dimensions at once and recovered 340,000 patient records.

How do you stop data quality degrading again after remediation?

Monitor the dimensions, not the score. St. David's now has automated checks across all ten quality dimensions running daily, alerting the data team within 24 hours of any falling below threshold. The previous fault ran undetected for a year and a half.