I am enjoying that all-too-brief time between finishing one big project and starting the next. It’s a time to read some emails I might have skipped and even take a minute to ponder them.
One email recommmendedThe Watch for Flu and More. It is a well-written article for the public describing the benefits and challenges of syndromic surveillance for adverse reactions during H1N1 immunization in the U.S. “Harvard Medical School scientists are linking large insurance databases that cover as many as 50 million people with vaccination registries around the country for real-time [emphasis added] checks of whether people see a doctor in the weeks after a flu shot and why.”
This program challenged by three factors.
- The time-lag associated with respect to detecting rare events in a fast-moving vaccination program. I inserted the snarky “emphasis added” to point out that it is difficult to put “real-time” and “billing data” in the same sentence.
- Subject identification. For populations of 50 million, what are the chances in a large program like this that the patient demographic information will match between insurance and registry data? The good news is that occasional mismatches are tolerable in a statistical study. The bad news is that the goals of the study are to detect relatively infrequent adverse effects, so the error level will reduce the sensitivity of detection. Perhaps the most important outcome of this study will be determining whether it is possible to detect adverse events this way. If there is an adverse event and it is not detected soon, perhaps ethicists who speak about the balance between protecting privacy and achieving the benefit of healthcare IT will have a real-world example to show the value of a universal healthcare ID.
- Registry compliance. For many patients getting the flu vaccine is a cash transaction. It is hard to imagine that Safeway and the airport immunization clinics are registering all their clients. Where they are registering there is little incentive to collect identifying information as thoroughly as when a bill must be submitted.
Changing the Odds
It is really important that the Harvard study happen. Perchance it will be a major discovery on adverse event detection and registries. More likely it will sharpen the policy discourse on where population studies have value and what can be done to improve their sensitivity. It is unrealistic to depend on the proliferation of EHRs into doctors’ offices to be complete in, say, 10 years. What is the level of proliferation that makes direct transmission of clinical data including immunizatin status a possibility for adverse event detection? Gartner believes that the combinaiton of incentive funds and Stark safe-harbor programs will create a pretty substantial nudge on proliferations. We could see as many as 20% of physicians using EHRs well enough to qualify for “meaningful use” by 2013. If those physicians are collecting immunization status and other indicators for adverse event reporting then the identity correlation would be reduced. If they have true “real-time” feeds of clinical data to massive adverse-event servers then the time-lags could be substantially reduced.
Being able to go beyond detection to associating problems with lot numbers will remain a challenge since patients will frequently get vaccinated through a different provider than the one that they use when symptoms develop.
Thinking in a ten-year timeframe it is also possible to think about policy changes that would motivate better use of data for various kinds of population studies. Perhaps it is possible to require those that adminster vaccines to register patients and submit the info to registries? It probably wouldn’t be cost-effective to pay them to register patients, given the limited value of adverse event detection for vaccines, on the average.
Absent a universal patient ID perhaps it would prove cost effective to pay PHR operators for providing linked up data sent with patient consent. PHR operators would likely use most of that payment to find ways to motivate patients to provide consent and authorize the necessary linkages with their providers. I’d bet these folks, who have a good handle on consumer behavior, could deliver a substantial number of patients for five bucks a head, and bask in the glory of making a proven contribution to population healthcare. Such a program might be cost effective because the data could be put to many uses. Like the Harvard study, a demonstration program on this model would yield important findings, succeeding or failing. It would provide some calibration of how much benefit a consumer must receive to share data. If it’s really $5 per person, or anywhere close, that would certainly inform the debate about the relative value that a consumer must receive to overcome privacy concerns.
Finally, simply requiring all health plans to play for all immunizations would create a data stream correlated on patient ID for analysis.
When it comes to achieving population-based value from clinical data, the hell of it is that we are required to accomplish this without a universal patient id. The good thing is that we are following Winston Churchill’s maxim, “when you are going through hell, keep going.” This Harvard study puts us one step forward.