Modelling evolving clinical practice guidelines: a case of Malawi

Abstract

Electronic medical record (EMR) systems are increasingly being adopted in low- and middle-income countries. This provides an opportunity to support task-shifted health workers with guideline-based clinical decision support to improve the quality of healthcare delivery. However, the formalization of clinical practice guidelines (CPGs) into computer-interpretable guidelines (CIGs) for clinical decision support in such a setting is a very challenging task due to the evolving nature of CPGs and limited healthcare budgets. This study proposed that a CIG modelling language that considers CPG change requirements in their representation models could enable semi-automated support of CPG change operations thereby reducing the burden of maintaining CIGs. Characteristics of CPG changes were investigated to elucidate CPG change requirements using CPG documents from Malawi where EMR systems are routinely used. Thereafter, a model-driven engineering approach was taken to design a CIG modelling framework that has a novel domain-specific modelling language called FCIG for the modelling of evolving CIGs. The CIG modelling framework was implemented using the Xtext framework. The national antiretroviral therapy EMR system for Malawi was extended into a prototype with FCIG support for experimentation. Further studies were conducted with CIG modellers. The evaluations were conducted to answer the following research questions: i) What are the CPG change requirements for modelling an evolving CIG? ii) Can a model-driven engineering approach adequately support the modelling of an evolving CIG? iii) What is the effect of modelling an evolving CIG using FCIG in comparison with the Health Level Seven (HL7) standard for modelling CIGs? Data was collected using questionnaires, logs and observations. The results indicated that finegrained components of a CPG are affected by CPG changes and that those components are not included explicitly in current executable CIG language models. The results also showed that by including explicit semantics for elements that are affected by CPG changes in a language model, smart-editing features for supporting CPG change operations can be enabled in a language-aware code editor. The results further showed that both experienced and CIG modellers perceived FCIG as highly usable. Furthermore, the results suggested that FCIG performs significantly better at CIG modelling tasks as compared to the HL7 standard, Arden Syntax. This study provides empirical evidence that a model-driven engineering approach to clinical guideline formalization supports the authoring and maintenance of evolving CIGs to provide up-to-date clinical decision support in low- and middle-income countries.