Domain-Specific Language Design for Patient-Oriented Prescriptions

Abstract Medical errors in hospitals cause 2 to 4 million patient injuries and 440,000 deaths in US hospitals annually; this amounts to one-sixth of annual US deaths (James, 2013; Kohn et al., 1999). Large studies and systematic reviews have found that healthcare information technology fails to reliably reduce patient injuries or deaths (Garg et al., 2005; Linder et al., 2007; Romano and Stafford, 2011; Zhou et al., 2009). Despite extensive effort and the expenditure of billions of dollars, computerization has failed to solve this problem (Landrigan et al., 2010). The PIs attribute this failure to a pervasive misunderstanding of the nature of computation in healthcare. While there has been a computer technology transfer in healthcare, we await an intellectual transfer, in which software design, maintainence, and debugging unlock the full potential of computer science to improve healthcare. The previous work of the PIs provides support for the claim that a prescription is a program. This observation implies that knowledge from computer science has direct application to the improvement of healthcare performance. We have shown that good software design applied to a paper prescription markedly decreases mortality associated with opioids in hospitalized patients (Belknap et al., 2008). To adapt software engineering techniques to implement and maintain highly reliable electronic prescrip- tions, the PIs have been designing and building the Patient-Oriented Prescription Programming Language (POP-PL). Clinicians who write prescriptions face the same spectrum of problems as professional pro- grammers who write software. What is needed are tools and methods that will permit clinicians to write prescriptions and take advantage of computer science knowledge about how to write and maintain effec- tive program. We expect to build such tools and develop such methods through application of our work with development of a domain-specific language (DSL). A DSL for prescriptions should build on existing concepts, such as order sets, but permit specification of actions to be performed in real time and to handle contingencies as they arise. This would include a means of providing the expressive power of abstractions over medical tasks—and it should provide a smooth path for clinicians to take advantage of more abstraction as they become more comfortable with writing programs. The PIs have experience building growable languages but significant improvements are needed to pro- duce DSLs suitable for healthcare. Specifically, while text is a powerful vehicle for abstraction, clinicians have traditionally employed graphical notations so as to address human-factor vulnerabilities. As reliable human–computer interactions are essential to the performance of prescriptions, the proposal targets support for a language with both graphical and textual notations in an extensible programming-language framework. Key Words. Domain-specific languages; programming languages; medicine; prescriptions Intellectual Merit. The PIs expect this work to generate multiple new insights regarding the building and maintenance of domain-specific language and new general methods for developing high-reliability software that facilitates collaboration between machines and humans. Computer science has long been driven by application areas, and core computer science discoveries have often been made because of a need to solve a particular problem. Given the power of the tools the PIs use and the cross-experience of our team, the proposed work is a tractable challenge that will produce imp