A Few Thoughts On Modernizing Healthcare Research
A few weeks ago I met a colleague / friend of mine who was visiting Bengaluru from Boston for a week, and we happened to spend time discussing the future of healthcare research and research tools. Here a few thoughts from the discussion, that I wanted to share in this post.
At ConvergeHEALTH by Deloitte, we have been now working for almost 3 years on developing state-of-art research solutions, with an end-goal to automate the research process into a few clicks. Essentially this would mean taking the entire painful 3–9 month research cycle of publishing a paper based on retrospective data into a short process which does not last more than a few days. For all the data geeks in healthcare, wouldn’t that be an amazing tool-set to have at our disposal!
This might seem to be a utopian vision, because healthcare data, analytics and research are incredibly nuanced in nature. Every research question has a lot of associated pre-conditions around how the hypothesis is generated, cohorts are formed, confounding factors are eliminated, statistical tests are performed, and how the results are interpreted. There are issues around data quality and governance which would arise as we start moving the needle on research tools, e.g., how do you prevent research analysts from data dredging (going trial and error mode) until they receive results that are favoring their hypothesis. All these factors make research automation a bit of a distant vision.
However, this field has seen some definitive progress in the past decade in terms of:
- Data capture: The last decade has seen a growing emphasis of “Data standards” and “interoperability” of data across different systems and type of data capture mechanism. A lot of data standards such as ICD-9 & ICD-10 for diagnosis of diseases, CPT for procedures, LOINC for laboratory tests, CDISC for clinical trial data, etc. have been emerging and getting refined over the past few years. Also, mobile data interoperability is getting bigger with giants like Apple & Google working on their own internal “health data standards” to allow for storage, integration and interoperability within various apps on their platform. While there is still a challenge associated around presence of too many standards and lack of uniform system to capture data, there are many investments around the world in this direction, and this sets the foundation for eventual consolidation.
- Data storage & warehousing: Not only have seen movement in terms of data capture, but data storage has evolved significantly in past few years. The ability to take large amounts of data from different health systems e.g., out-patient, in-patient, billing system, prescriptions, pharmacy data, laboratory data, and so on, and unify all that into a single longitudinal patient story is no small achievement. However, we have seen significant progress in terms of how data storage & warehousing has evolved as a science, improving the ability to capture and store data, and also use it for analysis purposes in a most efficient manner. OMOP has been now leading the charge for a few years to consolidate the industry learnings on this front, and share it in form an open-source model that organizations can use as a starting point for building their own data warehouse.
- Analysis tools: Analysis tools are the final applications which help us leverage this data and make something out of it. The amount of progress that we have seen in terms of how the analysis and statistical tools have improved over past few years is phenomenal. SAS, SPSS, R, etc. have improved their tools in terms of both capabilities as well as the over analytics horsepower to support usage of large amounts of data to derive meaningful results. However, these tools are still generic in nature and we await advancement to get more healthcare specific tools which require minimal configuration to meet healthcare specific use cases.
As a result, I believe that the next wave of progress will be standardization across methods for analysis and a great degree of automation in the execution process. We will see a lot of evolution in what one would call as “content” in the software world, which would include:
- Definitions: Development of standard disease and outcome definitions that are acceptable for various research studies, and that are usable across various distinct datasets would be of high value as we evolve the research tools. This not only makes research process more efficient, but also makes research comparable across various studies, and lays down the foundation for a much nuanced meta-analysis in large sample size by combining data across studies.
- Models: Availability of canned models which can be directly leveraged or configured for the purpose of a particular research. A great example of already available such models are the tools available in hospitals to flag, monitor and predict readmissions within hospitals. There is a much bigger potential to generate such models across various research steps including propensity matching to minimize confounders in an analysis, identifying a set of diseases at the baseline to characterize a cohort, calculating the accuracy of an analysis, understanding the disease progression based on configurable end-points, etc.
- Algorithms: Availability of talented people who can write various predictive and other algorithms on healthcare data is scare, and a lot of times these people end up working on same problems within different institutions. Some organizations are already taking charge on attacking this situation by developing some pre-tested algorithms that can be plugged-in to various systems or datasets, configured based on the specific scientific questions and used to perform an analysis. A good example of this could be a drug-switch analysis algorithm that we have been working on recently, where in simplistic terms the end research can configure a dataset, identify the two or more drugs between which they need to understand drug switch patterns and adjust for the granularity in terms of time based on the disease area being targeted.
- Data quality tools & frameworks: A standard procedure in healthcare research is to perform an assessment on the quality of data and it’s appropriateness for the particular type of research. As we modernize research, I see a lot of standardization being brought into this through common frameworks and tools executing upon those frameworks. This will enable a research to know the quality of data within a particular source at a high level, but also contextualize it in terms of the type of research question that he/she wants to answer. E.g., a claims dataset with large sample size and high longitudinality would be a great data asset for performing a cost of care analysis, but would act deficient for a cost-impact analysis due to lack of clinical data to quantify impact
- Therapeutic area specific content: Also once more dimension which will make all the research tools more advanced and nuanced is the availability of content that is specific to various therapeutic areas e.g., availability of data quality assessment tools for oncology, or availability of patient journey assessment tools for acute diseases
A big challenge that we talked about earlier is around governance in terms of ensuring the right scientific rigor, preventing data dredging, and ensuring results are shared and interpreted in the right manner. However, if we glance into history of how science has evolved, once we have the pieces ready it’s always easier to get them in place and develop the right governance around it.
All these research advancements in the pipeline, and the integration of different pieces of the puzzle, make me believe more than ever before that the vision of research automation is getting nearer, and definitely within reach where we will be able to impact medical science meaningfully not only for the coming generation but also for our own generation.
Note: While I am a Scientific Advisor to ConvergeHEALTH by Deloitte and run a preventive health platform Remedy Social, the views in this article are my personal views based on various experiences and conversations around developing healthcare research and analytics tools
— — — — — — — — — — — — —
Originally published at www.linkedin.com.