Using Link Analysis to Plan the Healthcare System

September 13, 2010
189 Views

Link Analysis can be used to understand where patients go to receive healthcare treatment and to identify bottlenecks in service that must be addressed. It is an interesting form of datamining that involves tracking populations as they move from point to point in a system. This analysis requires an ability to create hierarchical buckets (population segments) so that analysis may focus on the ‘% of the population’ as it moves from point to point in a system.

Link Analysis can be used to understand where patients go to receive healthcare treatment and to identify bottlenecks in service that must be addressed. It is an interesting form of datamining that involves tracking populations as they move from point to point in a system. This analysis requires an ability to create hierarchical buckets (population segments) so that analysis may focus on the ‘% of the population’ as it moves from point to point in a system. At one point, I know that SPSS and SAS datamining software had modules that automated this datamining process, and I once used a DataMart (Analytix) with a ‘Segment Manager’ module that was great for hierarchical programming.

Every Canadian has a Health Card that he or she must present when visiting a family physician, healthcare specialist (eg, Neurosurgeon), MRI or X-Ray clinic, hospital, medical centre, or other healthcare provider treatment. Data warehouses store details of every visit including a unique identifier code for every healthcare provider who was involved in the treatment. The healthcare provider who made the referral, the date on which the referral was made, and who the patient is scheduled to see next are also captured. With this data, a patient may be tracked as he or she moves from one healthcare provider to the next in the system.

As an example, let’s assume that there are the following fields of information about healthcare visits for 100 patients in a given month:

– patient identifier (HEALTH CARD #)
– date/time stamp of visit (VISIT DATE/TIME)
– code of the healthcare provider visited (PROVIDER VISIT)
– reason for visit (VISIT REASON)
– code of healthcare provider to whom patient is referred (PROVIDER REFERRAL)

Data exploration reveals that one or more of the following healthcare providers (PROVIDER) were visited by at least one or more patient during the month in question:

– Hospital (1111)
– Medical Centre 1 (2345)
– Medical Centre 2 (2346)
– Physician 1 (0023)
– Physician 2 (0024)
– Laboratory 1 (3346)

A preliminary analysis of the raw visit data may indicate that several patients have visited more than one healthcare provider, while others may have visited multiple providers on the same day.

Before a Link Analysis may be conducted, the PROVIDER VISIT and PROVIDER REFERRAL variables must be transformed from row to column format for each patient. For example, ‘Medical Centre 1’ must change from being a value in the PROVIDER variable to a boolean variable, and multiple visits to Medical Centre 1 require matching variables to capture this activity. ‘Medical Centre 1a’ and ‘Medical Centre 1b’ variables will capture that a patient made two visits to Medical Centre 1, whether on the same day or on different days. It is critical that the VISIT DATE/TIME variable be used to order the multiple provider variables from first to last date/time of visit. The process must be followed for each of the other 5 healthcare provider values in the PROVIDER variable.

Once this datashaping has been completed, the newly-created boolean variables must be rolled-up to reflect the ‘% of the population’ as it moves through each healthcare provider for treatment. SAS and SPSS had a tool that would present results in the form of a web: a thicker line connecting two healthcare providers reflecting a higher ‘% of the population’ that traveled between them. Viewing Link Analysis results graphically might lead us to conclude that there are bottlenecks in the healthcare system or providers who must have their patient load reduced and shared withy others.