Background and challenge

Delivering a high quality of care has always been an important pillar within the hospitals. Mapping and monitoring this quality of care can be done via different channels, such as the various registries, healthcare inspection and hospital accreditation that are supported by the government, but hospitals are also working internally on patient-reported outcome measures (PROMS), patient-reported experience measures (PREMS), and the monitoring of adverse events.

In the use case below, we will focus on the detection of adverse events. Such detection is usually done by means of random, manual file checks that are performed by the quality department. In some cases, hospitals use standardized material, such as the IHI Global Trigger Tool for Measuring Adverse Events, but in other cases, checks are made for hospital specific KPIs based on the request from doctors, nurses, or management.

Hospitals increasingly want to be able to respond quickly to any avoidable complications. With the help of CTcue, manual file checks can be accelerated, and adverse events can be detected while the patient is still in the hospital. In this way, a short feedback loop can be created with doctors and nurses and an ever-improving quality of care can be strived for.

Approach and results

In the use case below, we will zoom in on postoperative pneumothorax as a preventable complication. Postoperative pneumothorax can occur after barotrauma from mechanical ventilation, spontaneously due to ruptured blebs, as a consequence of invasive catheter insertion, or due to breakdown of the bronchial stump.

Traditionally, the occurrence of a postoperative pneumothorax is mapped using the coding of the minimal hospital data (MHD). However, this coding has a delay of several weeks or months, which makes a rapid response by the quality department impossible. By using CTcue - a Natural Language Processing (NLP) tool that can access both structured and unstructured information from the Electronic Medical Record (EMR) - these adverse events can be detected much faster.

The use case was therefore split into 2 parts. In a first phase, a query was developed that looks back at the past years and aims to detect all patients with a postoperative pneumothorax. The results of this query were then compared with the results of the minimal hospital data and served primarily as a validation step.

For this use case we were looking for patients who

• Had a surgery
• And had a mention of a pneumothorax in a report within 10 days after the surgery

These results were then compared with the results of the minimal hospital data. For this we were looking for all patients with MHD diagnosis code J95812 or J95811. This analysis showed that many more patients with a postoperative pneumothorax were picked up via CTcue than was the case via the MHD coding. This can be explained by the fact that the MHD coding is limited to analyzing the discharge letter, while CTcue also searched in radiology reports, operation reports, etc. In the context of the current use case, it seemed important to keep the focus as broad as possible, and not to limit ourselves to only one report type.

After validation, we moved on to phase 2 of the project. Here, the same query was rolled out, but this time limited to hospitalized patients who do not yet have a discharge date and are therefore currently still in the hospital. Since a postoperative pneumothorax is not a frequent complication, this query usually results in the detection of 0 patients. However, by running the query daily or weekly, the quality department can react quickly and, for each detected patient, talk to the physician involved to see if this complication could have been avoided.

Conclusion

Traditionally, detecting adverse events is not always easy. Often these adverse events cannot be found clearly in the structured data, meaning detection must be done by means of random, manual file checks. By using CTcue, these limitations no longer apply. Searching through reports is much faster, and always includes all relevant patients instead of a random selection.

Moreover, adverse events can be detected quickly, even while the patient is still in the hospital. Such rapid detection allows for an open discussion with the doctor or nurse involved on whether the complication could have been avoided. Depending on the root-cause of the postoperative pneumothorax, different preventative measures can be set up. Occasionally, surgical elements will induce a postoperative pneumothorax which need to be discussed with the surgeon involved, but more often, mechanical ventilation or catheter insertion will be the root-cause. Preventative measures such as close monitoring of the mechanical pressure, or ultrasound-guided catheter insertion could help to reduce the incidence rate. In this way, hospitals can strive for an ever-improving quality of care.

In the above use case, we have focused on the postoperative pneumothorax as a preventable complication. However, the same methodology could be applied for several other adverse events, i.e. the detection of in-hospital stroke, detection of intra-operative blood transfusion, detection of leakage after colon surgery, etc.

Do you want to get started with this yourself, or do you have any questions?

Then be sure to contact us for more information.