Generating Discharge Summaries in the MIS and Using Speech Recognition Software

Markus Loos

Introduction

The Charité--University of Medicine Berlin, is the largest hospital in Europe. Its four campuses have about 2,300 beds. The 9,384 employees care for about 100,000 inpatients and about 250,000 ambulatory patients per year. Approximately 4,900 students are instructed in medicine and dentistry.

The Department of Urology at the Benjamin Franklin Campus has 64 beds, 4 radiology workstations, and an outpatient department.

Background

At the Benjamin Franklin Campus of the Charité, there is an MIS (SAP-R/3) as well as many departmental systems. Since the SAP implementation in 1999, some of the departmental systems started to deliver their data via interfaces into the SAP system. Today doctors and nurses can, for example, view laboratory and radiological data within their SAP-workstation. Thus, the MIS has become a portal for the patients' medical data (the administrative data have existed in the SAP workplace for some time). Having all of these data together in one place offered the opportunity to automatically include these data in the patient's discharge summary, making it much easier for the physician to access important findings instead of hunting through the patient record for them.

History

Physicians in the Department of Urology dictated their discharge summaries, and while typing them, the secretaries used a Word template that imported the addresses of the general practitioner and urologist from a database.

For some years, the Head of the Department of Urology wanted discharge summaries to be created using speech recognition software. So in 1998, we first tested speech recognition software, after a "success story" in the German HIM's journal PMD ( Praxis Medizinische Dokumentation , German for "Medical Documentation in Practice"). This test failed. The reason was the lack of specific vocabulary, not the software itself. At the end of 1999, there was an article in PC Magazine that praised Dragon Naturally Speaking, so we gave it a try. While looking for a software dealer, we found a small software firm that sells the speech recognition software Dragon Naturally Speaking and develops specific vocabularies for medical purposes. This firm created a urological vocabulary based on different sources, including several thousands of discharge summaries written in the Urology Department at the Benjamin Franklin Campus of the Charité--University Medicine, Berlin.

Methods

SAP makes it possible to transfer medical data into Word documents, for example surgery reports or discharge summaries.

In an "SAP-only" setting, we use Word templates, which are filled with administrative and medical data, to create surgery reports and pre-written files for several surgical procedures. This allows name, date of birth, diagnoses, and procedures, with their corresponding ICD-10 and ICPM-OPS301 codes, to be automatically inserted into the surgical report as soon as the surgery is documented in the MIS. ICPM-OPS301 is the mandatory part for coding surgical procedures of the German ICPM translation. Depending on the surgical procedure performed, the urologist now selects a relevant surgical file that is copied into the surgery report. In this document, only minor changes (for example the amount of blood loss) have to be inserted to obtain a reliable surgery report for the procedure.

When drawing up discharge summaries, it is a little more complex. The physician identifies a patient and clicks "create document." As with the surgery report, some data are now automatically inserted into the document, but laboratory or radiological findings have to be manually inserted. Fortunately, these reports can be viewed completely within the MIS directly from the document creation screen. A conclusion of the radiological findings is documented at the end of the report. This conclusion is transferred via "cut and paste" into the discharge summary. To enable the input of handwritten or undocumented information, speech recognition software is used. Consequently, each urologist has his own headset and speaks directly to his computer. The speech recognition software uses a special urological vocabulary, which results in a good conversion of spoken language into written text. The physician corrects the text if necessary, using voice commands. The PCs chosen for using speech recognition software are not high-end workstations anymore. They need a 1GHz CPU, at least 512MB RAM, and a sound card with good input capabilities (Dolby 5.1 surround output is absolutely useless for dictating). The operation system is Windows NT. With Windows 2000 or XP, we could use USB-Headsets. Theoretically these headsets improve the input quality because the analogue-to-digital conversion takes place outside of the computer box and has no problems with internal noise.

Trends

During the old times of dictating by the physicians and typing by the secretaries, the urologists had no problems dictating data, which were already present in electronic form. This changed dramatically. While being forced now to correct any obscurities and mumbles themselves, they want as much data as they can get to be automatically imported into the documents. Getting the relevant information of radiological findings into the discharge summary is simple. Data from other departments (Laboratory or Pathology) are hard to get there. The Department of Pathology IT system doesn't send data to the MIS, and the Laboratory Systems drowned us in data. This forces us to dictate findings of the Departments of Pathology and Clinical Chemistry. We're now looking for an easy and efficient way to select and copy laboratory findings into the discharge summary.

Strategies

How do we get the laboratory data into the discharge summary? It's not a problem of a lack of data, but how to get relevant information out of masses of data. So automatic copying is no option, because for one parameter you need several findings, for others only the last value, for others again the highest (or lowest) value, and so on. Cut and paste is also no option, because we want only a part of the laboratory values in the discharge summary, not the whole finding. What we need is a complete overview of all laboratory values during a patient's stay in hospital. This overview needs an option to select single values for export into the discharge summary. Maybe it could be a screen form with a check box next to each value and an "export button" to bring every selected value at once into the Word document.

Results

The Department of Urology is now able to take almost all of what it needs out of the MIS. Depending on the diversity of the remaining information, either text modules for creating surgical reports or speech recognition software to build up discharge summaries is used. Thus the information from our department is easily gathered, but data from other departments are partially hard to get.

Conclusion

We have improved our methods of creating patient-centred medical documents. It's quite easy to get data from MIS into Word templates, for example surgery reports. There is still room for improvements, especially for seamless integration of findings from "external" IT sources into our discharge summaries.

References

Alwang, Greg. Speech Recognition. PC Magazine Vol. 18, 21(12/1/1999): 167-188.

Glatz, Volker. Erfahrungen mit einem Spracherkennungssystem. Praxis Medizinische Dokumentation Vol. 18, 3(1998): 41-45.