[ad_1]
Working with noise in the data is one of the challenges when dealing with free-text formatted input data. Filtering the input sentences and selecting more informative ones work as a solution to reduce the noise in the data when dealing with free-text categorisation of CFs. Finding a balance between removing the noise and keeping a sufficient number of features to have a well-trained model is a critical task.
The information-rich ngrams were selected according to their χ2 values, indicating the association between ngrams and a CF. The most relevant term for communication/hand-off failure in our cohort was fall followed by nurse and order. Glucose readings were repeatedly identified as an important contributor to policy/procedure issues. Our data’s selected information-rich ngrams for technology issues present the same trend through identifying medconnect and system as the most relevant ngrams for this CF. Besides, CGI, the stemmed form of Centigray (CGY), was the top information-rich ngram for technology issues. CGY is a measurement of absorbed radiation. This result suggests for technology-related CFs associated with radiation treatments such as with cyberknife radiation treatments. The ngrams with high χ2 values associated with lapse/slip were dose, pharmacy and order. The pattern among these ngrams indicates that the medication-related tasks were more prone to be affected by this CF. Our analysis indicated that the data entry process was affected by distractions/interruptions as some of the information-rich ngrams for this CF were data entry and error. Box 1 presents examples of information-rich ngrams, which identified information-rich sentences in PSE reports.
Instances of information-rich sentence selection algorithm applied on patient safety event (PSE) reports from different contributing factor categories.
Contributing factor
Communication/hand-off failure.
PSE report (brief factual description)
Patient was taken to nuclear medicine via transport for a scheduled stress test. Once he got to NM, the test was cancelled. Patient had drunk coffee with his breakfast because there was no NPO order in place for the test.
Information-rich Ngram
Selected information-rich sentence
Patient had drunk coffee with his breakfast because there was no NPO order in place for the test.
Contributing factor
Policy/procedure issue
PSE report (brief factual description)
A glucose test was performed at (time stamp 1) on patient by (nures 1) with a result of 36 mg/dL. The test was performed at (time stamp 2) by (nurse 2) with a result of 139 mg/dL, which was 1 hour and 4 min later. The Hypoglycaemia Policy states that a patient with a glucose less than 40 mg/dL should be treated and a glucose run every 15 min until the glucose returns to 90 mg/dL.
Information-rich Ngram
Selected information-rich sentence
A glucose test was performed at (time stamp 1) on patient by (nures 1) with a result of 36 mg/dL. The Hypoglycaemia Policy states that a patient with a glucose less than 40 mg/dL should be treated and a glucose run every 15 min until the glucose returns to 90 mg/dL.
Contributing factor
Technology issue.
PSE report (brief factual description)
I was unable to gain access to pyxis. Rebooted system and tried several interventions but unsuccessful. ICU and ED called to report inability to gain access to pyxis. Carefusion called and stated that the database was disconnected from the system and unable to diagnose problem at this time. Instructed to call help line and high priority ticket initiated. Patient began seizing. Medication system down and unable to obtain ativan in the ED. Nurse had to physically go to the pharmacy to obtain medicine.
Information-rich Ngram
Selected information-rich sentence
Rebooted system and tried several interventions but unsuccessful. Carefusion called and stated that the database was disconnected from the system and unable to diagnose problem at this time. Medication system down and unable to obtain ativan in the ED.
Contributing factor
Lapse/slip.
PSE report (brief factual description)
Order for an HIV med entered on the wrong patient. The pharmacists did not quesiton why the patient was ordered for only one HIV medication. The doctor called one afternoon asking how did this mistake happen and not be caught. At that time, that is when the pharmacists was made aware of the mistake.
Information-rich Ngram
Pharmacist.
Selected information-rich sentence
The pharmacists did not quesiton why the patient was ordered for only one HIV medication. At that time, that is when the pharmacists was made aware of the mistake.
Contributing factor
Distractions/interruptions.
PSE report (brief factual description)
Three prescriprions were e-scribed for one of our long-term patients here at store #N. All prescriptions were prepared and dispensed expediciously since our client was in a hurry to make his ride. All the medications were controlled except for one medication. The next day, we received a call from the doctor’s office, which happens to be a first time doctor for this client, stating that one of the medications were to be dispensed at a later date on ((date)). Unfortunately, missed that date at data entry as I performed the data entry of the prescriptions. The team did contact the patient and informed him of the prescriber’s specific directions in regards to that one prescription.
Information-rich Ngram
Selected information-rich sentence
Unfortunately, missed that date at data entry as I performed the data entry of the prescriptions.
We included unigrams and bigrams in the bag-of-words to calculate their χ2 value. Information-rich unigrams were more common than bigrams, perhaps because unigrams contain more generalisable information. However, bigrams can convey more specific information, but they are susceptible to noise. Further investigation is needed to assess the effect of bigrams in the information-rich sentence selection algorithm.
[ad_2]
Source link
