Developing a Quality Audit Report for General Practice Prescribing for Hypertension- Methodology

• Abstract
• 1. Introduction
• 2. Methodology
  • 2.1 Setting / Expert Panel
  • 2.2 Data Extraction
  • 2.3 The Quality Audit Report
  • 2.4 Validation
• 3. Preliminary Results
• 4. Discussion
• 5. Acknowledgements
• 6. References
• Appendix

Abstract
This research aims to design and validate a Quality Audit Report (QAR) for clinical audit in general practice chronic disease management. We focus on the suitability of routinely collected Patient Management System (PMS) data as the basis for quality audit statistics. We are working with an expert panel representing the clinicians of a general practice to identify appropriate PMS-derived statistics to provide descriptive, supportive and cautionary characterisations of their chronic disease management. The report development is centring on their management of hypertension and relevant co-morbidities. After several iterations of report development a sample of patients will be reviewed by the expert panel, blind to the patients’ reporting classifications, to assess the sensitivity and specificity of the QAR. Lessons from initial iterations of the QAR development process indicate that the temporal subtlety and explosion of possible reporting statistics are major issues in defining useful quality improvement statistics where contraindications and combination therapies are involved.
 

1. Introduction
This research aims to design and validate a Quality Audit Report (QAR) and lays the foundation for its use as a quality improvement process with the purpose of better chronic disease management. The hypothesis is that QARs, based on routinely collected Patient Management System (PMS) data, are a useful determinant of the quality of patient care with respect to prescribing for cardiovascular disease (CVD) risk management.

With CVD being the number one killer in New Zealand,^[1] there is a motivation to improve management of risk factors for CVD (eg, high cholesterol, obesity, and hypertension). General practitioners (GPs) are at the forefront of primary care for chronic disease management in terms of prescriptions and counselling for risk prevention. For high-quality patient care, it becomes essential that the health care providers offer the appropriate care to the patient based on the evidence-based clinical practice guidelines (doing the right thing) and in a safe and timely fashion (doing it right).^[2]

The presence of significant routinely collected data in PMSs encourages investigation of their utility to support quality improvement. Electronic medical record (EMR) systems have been successfully adopted in primary care in the Netherlands^[3] and in Australia.^[4] General practices in New Zealand have a near 100 percent computerisation rate with 89.7 percent of GPs using PMS for prescriptions.^[5] Stored data in the EMR systems are considered to be of high quality as a record of prescribing (sensitivity has been found to be near 100%),^[6] and, thus, provide a suitable basis for quality assurance and clinical audit reporting. Clinical audit is a process for quality improvement that seeks to improve patient care and outcomes through systematic review of care against explicit criteria and the implementation of change.^[7] To achieve such a "systematic review", the "explicit criteria" for quality audit must be agreed in advance.

This research attempts to lay the foundation for use of a QAR to support a quality improvement process for chronic disease management by general practices. The QAR will be comprised of various quality indicators, classified into descriptive, supportive and cautionary, for chronic disease management. We hypothesise that quality indicators based on changes in a patient’s therapy over time, as indicated by routinely collected information in the PMS, can be utilised for clinical audit. We will develop specific quality indicators in consultation with an expert panel of clinicians from an Auckland-based general practice. We will then validate the QAR by applying it to longitudinal electronic prescribing data extracted from the practice’s PMS and quantify the limitations of this method into sensitivity and specificity following review of cases identified in the report by the expert panel.

The current paper details the methodology for this research, provides some preliminary findings from the expert panel process (which is still in an early iteration) and discusses implications. This research is being conducted under University of Auckland Human Participants Ethics Committee protocol number 2007/078.


 

2. Methodology
 

2.1 Setting / Expert Panel
An expert panel of five members has been formulated by the participating general practice and consists of members of its clinical staff (two GPs, one of whom [JK] is also one of the investigators; the practice manager; and two nursing staff). It is important to note that, while extracts of patient data are processed by the University-based researchers, no identifying data (no data outside of a set of clinical observations and events, see below) is shared with anyone who is not already engaged in the care of those patients. The study will not directly entail any intervention on patients, however the clinical staff of the participating practice may well wish to adjust the treatment of patients where (and if) non-optimal treatment is pointed out by the analysis of their own PMS data.
 

2.2 Data Extraction
We have extracted de-identified electronic prescribing data from the PMS in use at the Auckland area general practice. The expert panel will be able to identify the patients for further follow-up on their cases while the researchers conducting data analysis will be blind to patient identity. The data extract will contain all patient information in the Practice’s PMS, within a selected date range, on:

1. Demographics (age in years, gender, ethnicity as coded in the PMS)
2. Dates of encounters
3. Classifications/problems
4. Prescriptions
5. Blood pressure measurements
6. Laboratory measurements (screening test results held in the PMS, as agreed between the investigators and the expert panel).

The data extract spans backwards from the time of extraction to 18 months prior, with the exception of Classifications/Problems, which are relevant for an indefinite time with respect to chronic illness and, hence, are extracted for five years back. Medications from the prescriptions are grouped under three therapeutic categories – antihypertensive agents, hypolipidaemic agents and hypoglycaemic agents – and with a variety of subcategories as dictated by the areas of interest identified by the expert panel. The therapeutic state-transition framework^[9] will be applied to derive the therapeutic states and state-transitions of the patients during their management.
 

2.3 The Quality Audit Report
A prototype of the QAR for prescribing to control CVD risk factors will be developed that will contain three categories of quality indicators:

1. Descriptive: These give a general description of the extracted electronic health records from the PMS
2. Supportive: These support the practice performance with an eye to quality assurance (relatively high numbers are indicators of good performance)
3. Cautionary: These instruct about the alignment of the practice to clinical practice guidelines and best practice (cases registering in these categories are recommended for review).

At the time of this writing a first draft of the QAR has been prepared for feedback from the expert panel (see Preliminary Results below).

Clinical practice guidelines selected for this study are those which are applicable to the practice in this research, are evidence-based and current. A range of candidate guidelines^[8-11] has been discussed with the panel. The expert panel will review the acceptability, applicability and correctness of the quality indicators developed by the researchers based on agreed guidelines and the priorities and areas of interest identified in discussion with the expert panel.
 

2.4 Validation
After no more than four meetings wherein revised versions of the QAR will be considered, the reporting criteria will be finalised and applied to the PMS’s extracted data. The finalised QAR will report agreed relevant characteristics of chronic disease management prescribing. While the initial focus has been CVD risk management, the concerns of the expert panel have brought forward Renal Failure as another key issue for their (largely Pacific) population. The result is tending toward a QAR that will focus on antihypertensive prescribing with consideration of key co-morbidities.

To evaluate the quality of the reporting, 20 individual de-identified cases will be randomly based on the criteria in the QAR in each of the following categories:

• Patients in a cautionary category (eg, with an apparent contraindicated use of beta-blocker due to asthma appearing on their problem list in the PMS).
• Patients under treatment to control any of hypertension, lipid or glycaemic status but that have NOT matched to any cautionary category in the QAR.

The expert panel will review each of these 40 cases, blind to whether the case has been matched to one or more QAR cautionary criteria or not. The review protocol will involve review by the clinical staff of the PMS data and any hardcopy files that are maintained and completion of the review form as per the Appendix. The resulting classifications will allow assessment (albeit with a somewhat broad confidence interval) of the sensitivity and specificity of the QAR cautionary categories as an indicator of high-quality chronic disease management within the scope of criteria considered.

Past experience with a similar protocol in Australia indicates a variety of sources of error (or providing cautions that turn out to be of minimal clinical utility) intrinsic in the method:^[12]

• Known non-compliance with the doctors’ advice
• Using supply of medications from another source – another general practice, existing oversupply, samples of new medications
• Personal intolerance of specific drug group that might otherwise be indicated for the patient
• Technical misinterpretations (an error in the analysis logic or the electronic health record not reflecting the full story).

Quantification of these errors will be of use to understand the potential (and limitations) of quality assurance use of the PMS database, as well as to document how improvements in national health IT infrastructure could improve the feasibility of such quality assurance efforts. Satisfactory outcomes will lay the foundation for use of the QAR in a quality improvement initiative.
 

3. Preliminary Results
Table 1 shows a summary with examples for the types of QAR statistics that arise from discussion with the expert panel. Several issues emerge as worthy of further consideration.

1.Outcome v Process. Actually, we are not dealing with "true" outcome measures, since these would be CVD events or death, and the closest we are inclined to come to outcomes is known correlates of outcome such as controlled BP. Other measures that may be considered outcomes for our purposes can be the presence of specific evidence based treatments (eg, due to conditions such as myocardial infarction, possibly related to overall CVD risk). Avoidance of contraindications occupies a middle ground between outcome and process. Further along the outcome/process continuum are measures that relate to treatment process, such as percentage of hypertensive patients on combination therapy, and the presence of appropriate and timely monitoring, irrespective of the levels found.

Table 1 – Examples of types of QAR statistics for patients classified with hypertension

2. Temporal issues. Time becomes a subtle issue in the definition of the quality audit criteria. Consider figure 1. There are at least four semantically distinct and interesting temporal relationships between a treatment via a prescribed drug and the classification of a patient to a chronic illness where that drug is contraindicated. Moreover, since we are considering a quality audit report, such reporting typically concerns a period of investigation with a start and end time of its own, which further multiplies the possible semantic categories. Consider, for instance, case C from figure 1. For Investigation Period 1, the contraindication is committed during the reporting period and is ongoing at the end of the reporting period – thus, it is not only a candidate to contribute to a cautionary criterion, but also is a candidate for immediate action (eg, to recall the patient or otherwise take action to change the medication regimen). For Investigation Period 2, however, the same event has been corrected before the end of the reporting period. Since it had been happening during the reporting period it could still contribute to a cautionary statistic; however, the repair of the condition means that it is no longer an item for present action, and the presence of the correction in the reporting period could even be considered a candidate for a supportive statistic. Further, and more complex, temporal issues arise from combination therapy and from the need to confirm conditions, such as hypertension, with multiple observations.

Figure 1 - Timeline illustrating temporal relationships amongst treatment, classification and reporting period

3. Explosion of options. Temporal issues account in part for this final problem – there are simply too many candidate statistics to report. Time for quality audit is limited (time spent in reflection must be balanced against time spent in clinical practice). It is impractical to aspire to a "complete" set of statistics that describe every temporal nuance of the relationships amongst all relevant clinical concepts. QAR statistics will be only a set of guideposts representing a much larger universe of possible measures. For this reason, we believe the input and buy-in of the practice staff is vitally important to achieve a locally relevant and motivating set of statistics.
 

4. Discussion
We are entering an exciting era where general practices now have the data, and are on the verge of having the tools, to conduct routine clinical audit for improvement of prescribing and chronic disease management outcomes. We are in the processes of developing and validating a QAR with a general practice that focuses on antihypertensive prescribing based on data readily extractable from the practice’s PMS. It should be said that there is some non-trivial complexity in data preparation and the deriving of clinically meaningful concepts such as combination therapy and problem-drug contraindications (relevant models are discussed at greater length in Gadzhanova 2007^[12]). However, these are within the realm of possible enhancements to future reporting tools that may be embedded in the PMS itself.

While we are currently focusing on the process of clinical audit within a single practice, and with locally defined and agreed reporting criteria, the concepts around aggregate analysis (for a group of practices, a region, or nationally) are not unrelated. CBG’s HealthStat (see http://www.healthstat.co.nz) demonstrates the potential for similar reporting based on nation-wide sampling of PMS data on short reporting cycles (eg, weekly). In either case there are a number of issues worthy of careful consideration:

• The temporal subtlety of data must be thought through – there are many possible relationships amongst clinical events and reporting periods and we must be clear about what we ask for and what interpretation is warranted.
• The immensity of possible options – for comparison purposes, and to motivate improvement, there is merit in choosing a set of national performance indicators; however, we should also encourage the development of local indicators, both to recognise local priorities and to reward the search for superior quality indicators (and what is superior will, of course, change in a dynamic environment).
• The need to validate the data – in Australia it was found that alert criteria from PMS data had good specificity but limited sensitivity when compared to a manual review of cases.^[12] The current research will provide another iteration of validation, and it should be acknowledged that the results are likely to uncover some limitations in how well a practice can assess its performance from the PMS data alone.

A further interesting issue in the sustainability of QAR efforts is the question of roles and rewards for such analysis. Better tools may reduce the effort in producing quality reports, but they still require analysis and subsequent action. Is this role to be borne by the GP as a professional re-accreditation requirement? Should it be an expansion of the practice manager role? How will the effort be funded?
 

5. Acknowledgements
The authors acknowledge the support and participation of Pasifika Health Care as essential to this research. We thank MedTech New Zealand for provision of research licences of MedTech32 to The University of Auckland.
 

References
 

1. Turley M, Stefanogiannis N, Tobias M, van der Hoorn S, Lawes C, Ni Mhurchu C, Rodgers A. Nutrition and the burden of disease: New Zealand 1997–2011. Public Health Intelligence Occasional Bulletin Number 17. Wellington, Ministry of Health; August 2003.
2. Seddon M. Quality improvement in healthcare in New Zealand. Part 1: what would a high-quality healthcare system look like? N Z Med J 2006;119(1237):U2056.
3. Bates DW, Ebell M, Gotlieb E, Zapp J, Mullins HC. A proposal for electronic medical records in US primary care. J Am Med Inform Assoc 2003;10(1):1–10.
4. McInnes DK, Saltman DC, Kidd MR. General practitioners’ use of computers for prescribing and electronic health records: results from a national survey. Med J Aust 2006;185(2):88–91.
5. Didham R, Martin I, Wood R, Harrison K. Information technology systems in general practice medicine in New Zealand. N Z Med J 2004;117(1198):U977.
6. Thiru K, Hassey A, Sullivan F. Systematic review of scope and quality of electronic patient record data in primary care. Br Med J 2003;326(7398):1070.
7. Baker R, Stevenson K, Shaw E, Redsell S. Principles for best practice in clinical audit: National Institute for Clinical Excellence. In: Radcliffe Medical Press; 2002.
8. Therapeutic Guidelines Ltd. Therapeutic guidelines: cardiovascular. Version 4. Therapeutic Guidelines Ltd, Australia; 2003.
9. Texas Department of State Health Services. Hypertension algorithm for diabetes mellitus in adults, Revised 1-26-06. Available via the Internet: http://www.dshs.state.tx.us/diabetes/PDF/ algorithms/ HYPER.PDF. Accessed July 2006.
10. Chobanian AV, Bakris GL, Black HR, Cushman WC, Green LA, Izzo JL Jr, Jones DW, Materson BJ, Oparil S, Wright JT Jr, and Roccella EJ. Seventh report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure. Hypertension. 2003 Dec 1;42(6):1206–52.
11. The National Heart Foundation of Australia. Hypertension management guide for doctors 2004. Available via the Internet:http://www.heartfoundation.com.au/downloads/hypertension_management_ guide_2004.pdf. Accessed July 2006.
12. Gadzhanova S, Iankov, II, Warren JR, Stanek J, Misan GM, Baig Z, Ponte L. Developing high-specificity antihypertensive alerts by therapeutic state analysis of electronic prescribing records. J Am Med Inform Assoc. 2007 Jan–Feb;14(1):100-9.

Appendix