Automated Proof Reading of Clinical Notes

Automated Proof Reading of Clinical Notes Jon Patrick and Dung Nguyen School of Information Technology, University of Sydney, 1 Cleveland St, Sydney 2006, NSW, Australia {jonpat, nguyend}@it.usyd.edu.au Abstract. Misspellings, abbreviations and acronyms are very popular in clinical notes and can be an obstacle to high quality information extraction and classification. In addition, another important part of narrative reports is clinical scores and measurements as doctors infer a patient‟s status by analyzing them. We introduce a knowledge discovery process to resolve unknown tokens and convert scores and measures into a standard layout so as to improve the quality of semantic processing of the corpus. System performance is evaluated before and after an automatic proof reading process by comparing the computed SNOMED- CT codes to the coding created originally by the clinical staff. The automatic coding of the texts increased the coded content by 15% after the automatic correction process and the number of unique codes increased by 4.7%. Accuracy of the automatic coding and annotations in the notes which have not been coded by the clinical staff is suggested by the system output. Keywords: clinical, proof reading, spelling, normalization, standardization. 1 Introduction Clinical notes contain valuable information about patients‟ status, however, retrieving information from them is challenging because they may comprise up to 30% non-word tokens, idiosyncratic spellings, abbreviations and acronyms, and poor grammatical structure. Besides resolving misspellings, knowing the correct expansions of abbreviations and acronyms is critical to the understanding of the document for both automatic natural language understanding and human comprehension and interpretation (Pakhomov et al., 2005). Proof reading is a process whereby a clinical text is validated to identify unknown tokens/words and their valid forms. Proof correcting is modifying the proofed text to make it notionally “correct” text and thereby more readily processible by automatic means. There are two principal tasks to be achieved, these are normalization and standardization. The normalization process changes the texts in a way so that a human reader would consider it as normal, such as correcting spelling, expanding abbreviations and acronyms. The standardization process converts the text into certain formats that an expert community has defined as standard; a good example is converting scores and measures into a standard layout. At first glance it would seem that standardization should be done initially before normalization, however it is more likely that both will need to be performed multiple times in a repeated cycle of processing as there is interaction between the two processes. Standardization converts various instances of the same scores and measures into standard forms so that the system does not need to be concerned about their details for later processing steps such as normalization. For example: “HR 70” and “HR 78” are standardized to the representation of heart rate. However, most measurements and scores contain acronyms or abbreviations which may need to be expanded during the normalization process. Normalization could improve the standardization process by correcting misspelling and other token error within standard forms. If normalization is executed first, a large amount …

Content truncated to 500 words