A Path to Analytics

Everyone agrees that “analytics” are/will be important for healthcare organizations, especially organizations with limited &/or constrained resources such as community health centers (CHCs), acute care hospitals, many rural hospitals & clinics of all kinds.  These organizations are facing many changes currently & are in the process of evolving to new organizational (participants in HIEs, ACOs, HCCNs etc.) & sustainability (service providers, data providers) models. What this means & how to do it are hotly discussed topics, however, with no apparent tactic or strategy that seems feasible. There is no big bang in this effort. This capability will not spring forth complete & productive if such organizations make the correct invocation or even spend a large amount of money. There does, however appear to be a set of steps that could provide an actual path for healthcare organizations with limited resources to follow to begin to productively use analytics & to evolve a more & more effective capability in this area. For simplicity, I’ll use CHCs as the example as I know them best.

So first, let’s define our terms: what are analytics, and why are they important? In this context analytics are essentially the extensive use of data, statistical and quantitative analysis, explanatory and predictive modeling to evaluate healthcare performance. In the health center context, UDS is a form of analytics as are the calculations done for sliding scale determinations, analysis of administrative and demographic data to improve the effectiveness of scheduling, and the analysis of cost data. Analytics are important not only within the health center for planning, quality improvement & clinical programs, but to help the center evaluate and engage in activities in the broader healthcare landscape.  Analytics are important for planning community health initiatives and contributing to public health programs, and they are essential for participation in health insurance exchanges and ACOs. Finally, analytics applied to clinical data can lead to better diagnosis, more effective treatment, enhanced care coordination and improved outcomes. Analytics, simply, is the systematic analysis of data focused on answering a specific question. It’s not magic; many healthcare organizations are already doing (some of) it.

Analysis of analytic procedures used by health centers & other resource-constrained organizations show the following common elements:[1]

·      use of currently available data, primarily directly from the EHR , but also from data aggregates, often provided by PCAs, networks, HIEs etc.

·      use of relatively simple analysis and visualization techniques, mainly through the EHR or the database that stores the aggregates (usually MySQL, MS SQL Server or Oracle, but in some cases MS Excel)

·      the development of simple linear models, such as analysis of variance, analysis of covariance, various types of regression, etc.,  or other straightforward statistical models done through the database, through MS Excel or less often through an analysis package (like SAS) or a business intelligence tool

·      and, most importantly, asking the right questions, that is, ones that are important for administrative and clinical operations and planning, .

Complex analytics, multi-layered analytics and highly designed data warehouses may not be initially necessary, at least for many questions, and moreover, not appropriate if the questions that are asked aren’t relevant or don’t require them and the underlying data isn’t complete and reliable. Recent experience with a data warehouse & analytics effort performed by a PCA & a contracted entity to perform the analysis have shown some of the issues with this approach including:

o   needing to hand select & normalize data due to the lack of ETL (extract, transform, load) activity (with the inevitable subjective component)

o   use of cumbersome or inappropriate tools for data organization & analysis

o   lack of skill in analytic planning & performance[S1] 

These issues can be addressed, & these common elements can be transformed into a program to evolve a path to analytics that can be piloted & then followed by health centers. The primary aspects of  the proposed “Path to Analytics” program would be[S2] :

1.     Developing a landscape analysis:

1.1.   what is being done within the organization already & by CHCs, PCAs, HCCNs, ACOs, HIEs & organizations also by with extensive resources such as Kaiser Permanente, Partners Healthcare, The Mayo Clinic, Geisinger Health System etc.

1.2.   landscape analysis is essential for developing context & particularly to have a range of examples for what similar organizations are already doing & what more advanced organizations show is possible

1.3.   this analysis can be done as a dialog with other organizations & can surface potential collaboration efforts as well as analysis examples

2.     Initiate a training program for analysis planning & analysis operations: what question(s) to ask, how to address them analytically, what tools to use, setting goals & metrics

2.1.   this training is essential for developing the skills not just in performing analytics & interpreting results, but also in how to develop relevant analytic hypotheses

2.2.   much training material is available online, especially from open source vendors

2.3.   commercial vendors & consultants may also provide low-cost or free training for other considerations (use in advertising, recommendations for additional work, etc.)

3.     Perform analytics planning: what question(s) to ask? how to ask them? what tools are compatible with answering them? 

3.1.   Selection of infrastructure & analytic tools as well as consultants &/or vendor partners if appropriate

3.1.1.  infrastructure tools include data & information storage & management applications as well as import/export & normalization applications

3.1.2. analytic tools include actual analysis packages & front end applications for query building & presentation of results

3.1.3. organization & storage of data: evaluation of current storage mechanisms & analysis of movement to contemporary mechanisms (Hadoop, etc.), extraction of data from clinical & administrative systems (PM, EHR, Labs, Registries, etc.), ETL & normalization of data as necessary

3.1.3.1.substantial amounts of data/information in healthcare is stored in databases underlying clinical & administrative applications such as practice management, EHR, lab & registry applications. Most of these have relational databases as the underlying data storage medium. Decisions will have to be made whether to maintain this storage infrastructure or move to applications that are not datatype dependent & can manage & search very large amounts of data.

3.1.3.2.If relational or other structured mechanisms are maintained, extraction of data may include substantial ETL & normalization efforts.

3.2.   Contemporary analytic solutions that may not require data warehousing & ETL:

3.2.1.  Hadoop & Hadoop-like solutions (open source file systems)

3.2.1.1.Disco (standard file system, Python for query & reduce, Nokia Research)

3.2.1.2.Spark (in-memory query, UC Berkeley)

3.2.1.3.Storm (real-time, i.e. not batch like Hadoop, Twitter)

3.2.1.4.HPCC (separate ecosystem: Thor, Roxie, Pig,… LexisNexis)

3.2.1.5.Etc.

3.2.2. Select question(s) for initial analysis

3.2.3. Plan extraction (if necessary), query strategy, data analysis & visualization/reporting

3.2.4. Create & review analysis plan

4.     Perform analysis & interpretation: operation of analytic process

4.1.   carrying out the analytics plan, this may require several iterations of the analysis

5.     Begin outreach & communication: How to convey results? Who to tell? Communication process, etc.

The endstate of this plan will be as follows:

·      One  or more of the organization’s sites will have completed the full life cycle of an analytics project that will be strategic/productive for them.

·      Their personnel will have worked, potentially with an innovative vendor & expert consultants in this area, on the issue/problem identification, analytic planning, analytic performance, interpretation & evaluation & results dissemination phases of an analytics project.

·      This project will have been specifically designed to use current (& possibly appropriate vendor-contributed) resources, so that subsequent projects will be feasible & more easily done given this experience.

o   Appropriate resources in this sense are contributed resources that use an organization’s existing hardware & data management/aggregation capabilities (as opposed to the many analytics packages that require advanced hardware & data management capabilities, as well as highly trained personnel)

The intent of this plan is to allow a healthcare organization with constrained resources to carry out an analytics project (or more likely set of projects) that will set them up for the productive use of analytics in their everyday operations. Of course, this only works if it is part of a larger operational & strategic plan that has buy-in from both the top level of the organization & from the clinicians in the organization.

Analytics will be essential for healthcare organizations as they evolve to meet the challenges of new organizational & sustainability issues. Constrained organizations must be able to meet these challenges on their own terms in order to remain viable & beginning the analytics with what already exists, & adding a little more, will allow them to begin to develop this productive & strategic capability.

---

[1] Personal communications to D. Hartzband by a variety of CHC, PCA-based &/or other administrators & providers

---