Mining the Gems

Health Management Technology, Oct, 2001 by Homer Jr. Warner

Strategies to help CIOs effectively use clinical databases while avoiding the data warehouse money pit.

Since the 1960s, healthcare users have imagined large data stores from which they could mine nuggets of information that would improve business processes and patient outcomes. That dream of data mining from an integrated clinically oriented warehouse has yet to become a reality for several reasons:

* The cost of additional data collection and interfacing to feeder systems is high;

* Aggregate clinical data analysis is not a high priority and resources are assigned elsewhere;

* Key data is missing from the reporting database or, if available, is not normalized so it can't be compared;

* Data is available but is too cryptic for ad hoc queries from outside the IT department;

* The feedback period between data analysis, implementing process changes, and realizing outcomes improvements can sometimes take years to understand.

To build a useful warehouse, designers must have some idea of the demands future users may place on the database. By knowing the questions, the data warehouse team will know which data needs to be collected, what interfaces to other systems and databases need to be built, and how to model that data for efficient retrieval. But can anyone really predict future database demands?

The problem arises when the data you need is not in the database because no one thought to collect it or it is represented in a form that makes retrieval difficult or impossible.

Data warehouse projects have become money pits because the project teams have not had experience in data warehousing and ,they underestimate the effort or overestimate their skills, or the organization is not "data driven," or project demands are continually changing or expectations are too high.

Researchers, physicians and heathcare executives who suggest ways in which they would use the database in the beginning of a project often have completely different demands by the time the warehouse is ready for use months or even years after the project commences. It is the demands that require additional data collection or interfaces to disparate databases that can turn data warehouse projects into never-ending development projects that often run out of budget long before showing a return on investment (ROI).

A Market for Data Warehousing

In November 2000, 3M commissioned a consultancy to conduct interviews with 30 healthcare provider organizations to better understand how to turn clinical data into useful information in acute healthcare settings. The clients interviewed used database products from all major healthcare IT vendors. The interviews explored reasons for data warehouse successes and failures.

Comments from database architects and chief information officers (CIOs) ranged from sophisticated to basic:

* "We are getting at data to help us get to a six sigma performance level."

* "We have designed and developed high performance teams that use the warehouse data -- these operational teams look at everything."

* "We have the data; we just don't know how to use it."

* "We have lots of data, but no one is asking for it."

Most of those interviewed could not describe their organizations' goals for aggregate use of information. Many had dismantled efforts to unify data and had retrenched their efforts to get better data for future decision modeling. For many, missing data such as drug profiles were the biggest stumbling block to providing improved management reports. In fact, only 20 percent of those interviewed had all the key ancillary data to produce useful clinical reports with information blended from a variety of sources.

The consultants also contacted 116 vendors claiming to have either a clinical data repository (CDR) or data warehousing capability. Of the 116 calls:

1. Forty percent of respondents said that they had a clinical database but did not have a true clinical data warehouse that integrates data from across the enterprise for efficient aggregate analysis.

2. Thirteen percent provide data warehouse products and/or analytic services but the majority focus on financial and administrative rather than clinical data.

3. Eight percent have a CDR (to efficiently look up single patient data but not across patient populations) but no warehouse nor specialized analytic tools or data mining services. Crystal Reports, a third-party query tool, is the most commonly promoted product by healthcare IT vendors for querying their database.

4. Thirty-two percent failed to respond to the survey, and 7 percent had gone out of business.

From the vendor and provider interviews, the consultants developed a list of "currently available" reports most often generated by healthcare organizations (see Table 1) as well as a "wish list" of reports that healthcare organizations would like more often than they do now (see Table 2).

Table 1

High Priority Reports Often Produced by
Hospital Information Systems

Acute inpatient days: number of days of care for all acute
 inpatients
Acute inpatient discharges: number of acute, psychiatric and
 rehabilitation patients released
APR DRG by attending
Average cost per case
Average length of stay
Census, daily
Classification APG summary by payor
Classification APG summary by physician
DRG financial and management report
DRG with attending MD
Drug utilization by physician
Emergency department visits
Inpatient surgical procedures
Number of transfers per patient during a stay
Outpatient registration: number of patient registrations for
 various outpatient services
Outpatient surgical procedures
Patients receiving blood products
Physician profiles and volume indicators
Physicians delinquent in signing charts
Revenue by payor class
Revenue by product line
Total deliveries
Total inpatient discharges
Table 2

High Priority Reports Less Often Produced by Hospital Information
 Systems

Blood review-indicated status
CHF registry
Clinical pathways compliance
Clinical variance and total cost/case
Cost to treat a specific disease (using costs, not charges)
Cost trending
Disease management (e.g., where is facility losing money, why?)
Evidence-based protocols and protocol variance reporting
Medication errors
Medication use by outcomes
Medication use by physician
OB number of births by C-section
Physician/procedure/diagnosis comparisons (i.e., practice variations)
Procedures that may result in billing denials
Quality management indicators
Security (staff access to patient records)
Security (failed logins)
Severe adverse drug reactions
Surgery and OR costs by procedure
Trauma register
Users-per-hour report