PhD related

A bit about my PhD research:

Title:
ChronoMedIt – A Computational Quality Audit Framework for Better Management of Patients with Chronic Disease

Overview:
NZ general practitioners (GPs) are diligent users of clinical computing by world standards, using electronic records that include prescribing data, laboratory test results, patient problem classifications as well as observation data such as blood pressure measurements. These data provide relatively obvious capabilities for practice introspection or clinical audit. Simple reporting queries can answer questions such as how many angiotensin converting enzyme inhibitors (ACEi) were prescribed last quarter (actually, there may be a few tricks around accurately identifying all prescriptions in this class since drugs are usually prescribed using generic/brand names with no association to a ‘drug class’). It is only slightly more complex to identify the percentage of patients with hypertension and diabetes that were prescribed with ACEi within the last 90 days. What is harder is to identify what percentage of patients with hypertension and diabetes have a persistent coverage with ACEi throughout the last quarter with no therapeutically significant lapses.

This latter type of query has a strong “temporal” nature. Such queries can become even more demanding if we wish to examine the progression of blood pressure measurements, or evaluate therapy with respect to the timing of relevant laboratory observations. As part of this research, we worked with HealthWest Fono to formulate and validate clinical audit reporting capability around antihypertensive prescribing. The work identified the temporal query requirements for a reporting tool that can more readily support relevant questions apropos to chronic disease management. Taking the identified requirements as guidance on the nature of queries that need to be formulated, a model of chronic disease audit was developed with four broad classes of indicators: (1) persistence to indicated medication; (2) timely measurement recording; (3) time to achieve target; and (4) measurement contraindicating therapy.

The criteria model has been implemented with the ChronoMedIt framework (indicating Chronological Medical AudIt) as an extensible and configurable architecture. The main components of the ChronoMedIt architecture are: an XML based specification for indicator formulation (with an associated XML-Schema), a drug and classification knowledge base maintained using Semantic Web technologies, a C# based criteria processing engine, a SQL-Server based patient database with related stored procedures and a graphical user interface to formulate queries and generate reports. ChronoMedIt can produce patient-specific audit reports as well as reports to benchmark an entire practice for a given evaluation period. A visualisation tool has been developed to provide an alternate representation of patient prescribing and measurement histories. By modifying the indicator specification and knowledge base an analyst can address a wide array of chronic disease management queries as specific instances of the four broad indicator classes.  The framework’s core computation has been verified using redundant query implementations on a battery of simulated case data and is illustrated against the EMRs of several practices. A paper that discusses some of the computational challenges can be found here while the details of the proposed solution can be found in this paper.

We have applied the ChronoMedIt in several real-world settings already and shown below are some of the important findings based on patient EMR data from two general practices:

1. a significant portion (59% and 63% respectively for the two practices concerned) of patients with hypertension have >30 day lapses in their antihypertensive medication; and over a third of people with hypertension have not had a BP measurement for >180 days (see Pubmed);
2. at least 56% of patients with hypertension and diabetes showed poor adherence to ACEi/ARB therapy (MPR <80%), although as a result, these patients were more likely to have uncontrolled BP than adherent patients (odds ratio = 4.0, p = 0.002 and odds ratio = 2.5, p = 0.034 for the two practices) (see Pubmed);
3. adherence to antihypertensive therapy is correlated to having controlled BP with non-adherent patients being more likely to have uncontrolled BP than adherent patients (odds ratio = 2.4, p = 0.001 and odds ratio = 1.7, p = 0.03 for the two practices); mean reductions in systolic BPs were observed to be 19.31 mmHg and 16.39 mmHg respectively for the two practices for being adherent from 0% to 100% (see HINZ paper);
4. interval based measures, such as MPR, are more stable measures than single, point-in time measures in identifying patients with poor BP control;
5. satisfying a single, point-in time measure may not necessarily be an indication of optimal management of BP and other measures need to be considered, especially if quality indicators are associated with financial incentives (see Pubmed);
6. analysing prescribing data has much merit and can provide 81% PPV and 76% NPV for dispensing based non-adherence (Pubmed); and
7. 39% of the patients starting antidepressants were found to be non-adherent and it was shown that using the prescription-visualisation tool may provide an opportunity for clinicians to have more informed conversations with patients.

ChronoMedIt was used in a nurse-led medication adherence promotion feasibility study (funded by NZ Health Research Council) to identify and follow-up patients with poor antihypertensive medication adherence (see this paper for details). More recently, it was used to demonstrated widespread antihypertensive medication adherence problems in a Pacific-led general practice serving a predominantly Pacific (majority Samoan) caseload in suburban New Zealand (results published here – PubMed).