Surgery's Rosetta Stone: Natural language processing to predict discharge and readmission after general surgery.

BACKGROUND: This study aimed to examine the accuracy with which multiple natural language processing artificial intelligence models could predict discharge and readmissions after general surgery. METHODS: Natural language processing models were derived and validated to predict discharge within the next 48 hours and 7 days and readmission within 30 days (based on daily ward round notes and discharge summaries, respectively) for general surgery inpatients at 2 South Australian hospitals. Natural language processing models included logistic regression, artificial neural networks, and Bidirectional Encoder Representations from Transformers. RESULTS: For discharge prediction analyses, 14,690 admissions were included. For readmission prediction analyses, 12,457 patients were included. For prediction of discharge within 48 hours, derivation and validation data set area under the receiver operator characteristic curves were, respectively: 0.86 and 0.86 for Bidirectional Encoder Representations from Transformers, 0.82 and 0.81 for logistic regression, and 0.82 and 0.81 for artificial neural networks. For prediction of discharge within 7 days, derivation and validation data set area under the receiver operator characteristic curves were, respectively: 0.82 and 0.81 for Bidirectional Encoder Representations from Transformers, 0.75 and 0.72 for logistic regression, and 0.68 and 0.67 for artificial neural networks. For readmission prediction within 30 days, derivation and validation data set area under the receiver operator characteristic curves were, respectively: 0.55 and 0.59 for Bidirectional Encoder Representations from Transformers and 0.77 and 0.62 for logistic regression. CONCLUSION: Modern natural language processing models, particularly Bidirectional Encoder Representations from Transformers, can effectively and accurately identify general surgery patients who will be discharged in the next 48 hours. However, these approaches are less capable of identifying general surgery patients who will be discharged within the next 7 days or who will experience readmission within 30 days of discharge.

Systolic blood pressure levels and mortality in Australian medical inpatients.

The epidemiology of elevations in blood pressure is incompletely characterized, particularly in Australia. Given the lack of evidence regarding the frequency and the optimal management of in-hospital hypertension, the authors performed a multicenter retrospective cohort study of consecutive medical admissions in South Australia over a 2-year period to investigate systolic blood pressure levels and their association with in-hospital mortality. Among 16 896 inpatients, 76% had at least one systolic blood pressure reading of ≥140 mmHg and 11.7% of ≥180 mmHg during hospitalization. A statistically significant negative relationship was observed between having at least one reading ≥140 mmHg and a likelihood of in-hospital mortality (odds ratio 0.41, 95% CI: 0.35 to 0.49, P < .001). Our results suggest that elevations in systolic blood pressure are common in Australian medical inpatients. However, the inverse association observed between systolic blood pressure values ≥140 mmHg and in-hospital mortality warrants further research to determine the clinical significance and optimal management of blood pressure elevations in this group.

Quantifying time from last dose: do direct oral anticoagulant assays correlate with patient's reported last dose.

INTRODUCTION: In the absence of a patient's last direct oral anticoagulant (DOAC) dose time, best practice regarding preoperative DOAC cessation remains unclear. The aim of this study was to investigate, in a real-life patient cohort, if there was an association between subjective patient recall and objective DOAC assay titre. METHODS/MATERIALS: A multicentre cohort study of consecutive surgical inpatients was conducted. DOAC assays were 'expected' if they satisfied both time and titre-based guidelines. RESULTS: Patient-recalled last dose and DOAC assay was available in 285 individuals. DOAC assay titres correlated strongly with the expected levels based on a patient's reported last dose time(rho = 0.70, P value < 0.0001). However, underweight (<50 kg; P  = 0.0339) and elderly (>80 years; P  = 0.0134) were more likely to have an unexpectedly high assay titre. CONCLUSIONS: A significant portion (∼25%) of patients had unexpected DOAC titres. DOAC levels can be clinically impactful in a significant percentage of patients, particularly in elderly and/or underweight.

Sociocultural and Demographic Factors Predict Readmissions for General Surgery Patients.

INTRODUCTION: Readmission is a poor outcome for both patients and healthcare systems. The association of certain sociocultural and demographic characteristics with likelihood of readmission is uncertain in general surgical patients. METHOD: A multi-centre retrospective cohort study of consecutive unique individuals who survived to discharge during general surgical admissions was conducted. Sociocultural and demographic variables were evaluated alongside clinical parameters (considered both as raw values and their proportion of change in the 1-2 days prior to admission) for their association with 7 and 30 days readmission using logistic regression. RESULTS: There were 12,701 individuals included, with 304 (2.4%) individuals readmitted within 7 days, and 921 (7.3%) readmitted within 30 days. When incorporating absolute values of clinical parameters in the model, age was the only variable significantly associated with 7-day readmission, and primary language and presence of religion were the only variables significantly associated with 30-day readmission. When incorporating change in clinical parameters between the 1-2 days prior to discharge, primary language and religion were predictive of 30-day readmission. When controlling for changes in clinical parameters, only higher comorbidity burden (represented by higher Charlson comorbidity index score) was associated with increased likelihood of 30-day readmission. CONCLUSIONS: Sociocultural and demographic patient factors such as primary language, presence of religion, age, and comorbidity burden predict the likelihood of 7 and 30-day hospital readmission after general surgery. These findings support early implementation a postoperative care model that integrates all biopsychosocial domains across multiple disciplines of healthcare.

Vital signs and medical emergency response (MER) activation predict in-hospital mortality in general surgery patients: a study of 15 969 admissions.

BACKGROUND: The applicability of the vital signs prompting medical emergency response (MER) activation has not previously been examined specifically in a large general surgical cohort. This study aimed to characterize the distribution, and predictive performance, of four vital signs selected based on Australian guidelines (oxygen saturation, respiratory rate, systolic blood pressure and heart rate); with those of the MER activation criteria. METHODS: A retrospective cohort study was conducted including patients admitted under general surgical services of two hospitals in South Australia over 2 years. Likelihood ratios for patients meeting MER activation criteria, or a vital sign in the most extreme 1% for general surgery inpatients (<0.5th percentile or > 99.5th percentile), were calculated to predict in-hospital mortality. RESULTS: 15 969 inpatient admissions were included comprising 2 254 617 total vital sign observations. The 0.5th and 99.5th centile for heart rate was 48 and 133, systolic blood pressure 85 and 184, respiratory rate 10 and 31, and oxygen saturations 89% and 100%, respectively. MER activation criteria with the highest positive likelihood ratio for in-hospital mortality were heart rate ≤ 39 (37.65, 95% CI 27.71-49.51), respiratory rate ≥ 31 (15.79, 95% CI 12.82-19.07), and respiratory rate ≤ 7 (10.53, 95% CI 6.79-14.84). These MER activation criteria likelihood ratios were similar to those derived when applying a threshold of the most extreme 1% of vital signs. CONCLUSIONS: This study demonstrated that vital signs within Australian guidelines, and escalation to MER activation, appropriately predict in-hospital mortality in a large cohort of patients admitted to general surgical services in South Australia.

Beware of little expenses: Low-value endocrinological blood tests in geriatric medical inpatients.

OBJECTIVES: Blood tests for endocrinological derangements are frequently requested in general medical inpatients, in particular those in the older age group. Interrogation of these tests may present opportunities for healthcare savings. METHODS: This multicentre retrospective study over a 2.5-year period examined the frequency with which three common endocrinological investigations [thyroid stimulating hormone (TSH), HbA1c, 25-hydroxy Vitamin D3] were performed in this population, including the frequency of duplicate tests within a given admission, and the frequency of abnormal test results. The Medicare Benefits Schedule was used to calculate the cost associated with these tests. RESULTS: There were 28,564 individual admissions included in the study. Individuals ≥65 years old were the majority of inpatients in whom the selected tests were performed (80% of tests). TSH was performed in 6730 admissions, HbA1c was performed in 2259 admissions, and vitamin D levels were performed in 5632 admissions. There were 6114 vitamin D tests performed during the study period, of which 2911 (48%) returned outside the normal range. The cost associated with vitamin D level testing was $183,726. Over the study period, 8% of tests for TSH, HbA1c, and Vitamin D were duplicates (where a second test was performed within a single admission), which was associated with a cost of $32,134. CONCLUSIONS: Tests for common endocrinological abnormalities are associated with significant healthcare costs. Avenues by which future savings may be pursued include the investigation of strategies to reduce duplicate ordering and examining the rationale and guidelines associated with ordering tests such as vitamin D levels.

Why do we evaluate 30-day readmissions in general medicine? A historical perspective and contemporary data.

Reducing preventable readmissions is important to help manage current strains on healthcare systems. The metric of 30-day readmissions is commonly cited in discussions regarding this topic. While such thresholds have contemporary funding implications, the rationale for individual cut-off points is partially historical in nature. Through the examination of the basis for the analysis of 30-day readmissions, greater insight into the possible benefits and limitations of such a metric may be obtained.

Antibiotic Prescribing Practices Differ between Patients with Penicillin Intolerance and Penicillin Allergy Labels.

INTRODUCTION: Penicillin allergy labels are common. However, many penicillin allergy labels have been applied incorrectly and in fact represent penicillin intolerance. Patients with penicillin intolerance can receive penicillin antibiotics. The effect of penicillin intolerance labels on prescribing practices is uncertain. METHODS: This multicenter retrospective cohort study included consecutive general medicine patients admitted to two tertiary hospitals over a 12-month period. Electronic medical records were reviewed for allergy and prescribing practices. Instances of penicillin prescription to patients with previously labeled penicillin allergies underwent case note review. RESULTS: There were 12,134 individual hospital admissions included in the study. The number of admissions with a previous penicillin allergy label was 1,312 (10.8%) and with a penicillin intolerance label was 60 (0.5%). Penicillin allergy labels were associated with increased likelihood of being prescribed vancomycin (odds ratio 1.42, 95% confidence interval 1.16-1.75, p = 0.001) and moxifloxacin (odds ratio 20.0, 95% confidence interval 13.4-29.9, p < 0.001). Penicillin intolerance was not associated with increased likelihood of receiving these antibiotics. There were 75 admissions during which an individual with a penicillin allergy label was prescribed one of the specified penicillins and only one adverse reaction in this group. These cases included eight deliberate challenges and 15 cases in which allergy history clarification was sufficient to delabel the allergy. CONCLUSIONS: This study supports that prescribing practices differ between patients with penicillin allergy labels and intolerance labels. Penicillin challenges may be undertaken safely in the inpatient setting. Further studies are required to investigate how best to interrogate penicillin allergy labels in this cohort.

Muscle Protein Synthesis after Protein Administration in Critical Illness.

Rationale: Dietary protein may attenuate the muscle atrophy experienced by patients in the ICU, yet protein handling is poorly understood. Objectives: To quantify protein digestion and amino acid absorption and fasting and postprandial myofibrillar protein synthesis during critical illness. Methods: Fifteen mechanically ventilated adults (12 male; aged 50 ± 17 yr; body mass index, 27 ± 5 kg⋅m-2) and 10 healthy control subjects (6 male; 54 ± 23 yr; body mass index, 27 ± 4 kg⋅m-2) received a primed intravenous L-[ring-2H5]-phenylalanine, L-[3,5-2H2]-tyrosine, and L-[1-13C]-leucine infusion over 9.5 hours and a duodenal bolus of intrinsically labeled (L-[1-13C]-phenylalanine and L-[1-13C]-leucine) intact milk protein (20 g protein) over 60 minutes. Arterial blood and muscle samples were taken at baseline (fasting) and for 6 hours following duodenal protein administration. Data are mean ± SD, analyzed with two-way repeated measures ANOVA and independent samples t test. Measurements and Main Results: Fasting myofibrillar protein synthesis rates did not differ between ICU patients and healthy control subjects (0.023 ± 0.013% h-1 vs. 0.034 ± 0.016% h-1; P = 0.077). After protein administration, plasma amino acid availability did not differ between groups (ICU patients, 54.2 ± 9.1%, vs. healthy control subjects, 61.8 ± 13.1%; P = 0.12), and myofibrillar protein synthesis rates increased in both groups (0.028 ± 0.010% h-1 vs. 0.043 ± 0.018% h-1; main time effect P = 0.046; P-interaction = 0.584) with lower rates in ICU patients than in healthy control subjects (main group effect P = 0.001). Incorporation of protein-derived phenylalanine into myofibrillar protein was ∼60% lower in ICU patients (0.007 ± 0.007 mol percent excess vs. 0.017 ± 0.009 mol percent excess; P = 0.007). Conclusions: The capacity for critically ill patients to use ingested protein for muscle protein synthesis is markedly blunted despite relatively normal protein digestion and amino acid absorption.

Pre-hospital emergency anaesthesia in trauma patients: An observational study from a state-wide Australian pre-hospital and retrieval service.

OBJECTIVE: To assess the performance of an Australian pre-hospital and retrieval medicine (PHRM) service against the National Institute for Health and Care Excellence (NICE) standard which recommends that pre-hospital emergency anaesthesia (PHEA) in trauma patients should be conducted within 45-min of first contact with emergency services. METHODS: Retrospective observational study of all adult trauma patients in which PHEA was conducted by the PHRM service covering a 5-year period from January 2015 to December 2019. RESULTS: Over the 5-year study period, 1509 (22%) of the PHRM service workload comprised primary retrievals from scene. Most 1346 (89%) of these cases had a primary diagnosis of trauma. Of these we have complete data for 328 of the 337 cases requiring a PHEA and 121 (37%) patients received this within the recommended 45-min time frame. The service attended in rapid response vehicles (n = 160, 49%), rotary wing (n = 151, 46%) and fixed wing (n = 17, 5%) transport modalities. For a service covering 983 482 km2 , the median distance travelled to patients was 35 (16-71) km and the median time to PHEA was 54 (38-80) min. CONCLUSIONS: In a cohort of 337 patients treated by a dedicated PHRM service in South Australia, the median time to PHEA was 54 (38-80) min with only 37% of patients receiving PHEA within 45 min from the activation of the team. Despite differing patient demographics, the percentage of patients receiving PHEA within the recommended time frame was greater than a similar cohort from the UK. However, both data sets still fall short of recommended targets.

Mixed-mode versus paper surveys for patient-reported outcomes after critical illness: A randomised controlled trial.

OBJECTIVE: The aim of the study was to determine the response rate to a mixed-mode survey using email compared with that to a paper survey in survivors of critical illness. DESIGN: This is a prospective randomised controlled trial. SETTING: The study was conducted at a single-centre quaternary intensive care unit (ICU) in Adelaide, Australia. PARTICIPANTS: Study participants were patients admitted to the ICU for ≥48 h and discharged from the hospital. INTERVENTIONS: The participants were randomised to receive a survey by paper (via mail) or via online (via email, or if a non-email user, via a letter with a website address). Patients who did not respond to the initial survey received a reminder paper survey after 14 days. The survey included quality of life (EuroQol-5D-5L), anxiety and depression (Hospital Anxiety and Depression Scale), and post-traumatic symptom (Impact of Event Scale-Revised) assessment. MAIN OUTCOME MEASURES: Survey response rate, extent of survey completion, clinical outcomes at different time points after discharge, and survey cost analysis were the main outcome measures. Outcomes were stratified based on follow-up time after ICU discharge (3, 6, and 12 months). RESULTS: A total of 239 patients were randomised. The response rate was similar between the groups (mixed-mode: 78% [92/118 patients] vs. paper: 80% [97/121 patients], p = 0.751) and did not differ between time points of follow-up. Incomplete surveys were more prevalent in the paper group (10% vs 18%). The median EuroQol-5D-5L index value was 0.83 [0.71-0.92]. Depressive symptoms were reported by 25% of patients (46/187), anxiety symptoms were reported by 27% (50/187), and probable post-traumatic stress disorder was reported by 14% (25/184). Patient outcomes did not differ between the groups or time points of follow-up. The cost per reply was AU$ 16.60 (mixed-mode) vs AU$ 19.78 (paper). CONCLUSION: The response rate of a mixed-mode survey is similar to that of a paper survey and may provide modest cost savings.

Daily estimates of individual discharge likelihood with deep learning natural language processing in general medicine: a prospective and external validation study.

Machine learning, in particular deep learning, may be able to assist in the prediction of the length of stay and timing of discharge for individual patients. Artificial neural networks applied to medical text have previously shown promise in this area. In this study, a previously derived artificial neural network was applied to prospective and external validation datasets. In the prediction of discharge within the next 2 days, when the algorithm was applied to prospective and external datasets, the area under the receiver operator curve for this task were 0.78 and 0.74, respectively. The performance in the prediction of discharge within the next 7 days was more limited (area under the receiver operator curve 0.68 and 0.67). This study has shown that in prospective and external validation datasets the previously derived deep learning algorithms have demonstrated moderate performance in the prediction of which patients will be discharged within the next 2 days. Future studies may seek to further refine or evaluate the effect of the implementation of such algorithms.

The use of smartphone-derived location data to evaluate participation following critical illness: A pilot observational cohort study.

BACKGROUND: Disability is common following critical illness, impacting the quality of life of survivors, and is difficult to measure. 'Participation' can be quantified as involvement in life outside of their home requiring movement from their home to other locations. Participation restriction is a key element of disability, and following critical illness, participation may be diminished. It may be possible to quantify this change using pre-existing smartphone data. OBJECTIVES: The feasibility of extracting location data from smartphones of survivors of intensive care unit (ICU) admission and assessing participation, using location-based outcomes, during recovery from critical illness was evaluated. METHODS: Fifty consecutively admitted, consenting adult survivors of non-elective admission to ICU of greater than 48-h duration were recruited to a prospective observational cohort study where they were followed up at 3 and 6 months following discharge. The feasibility of extracting location data from survivors' smartphones and creating location-derived outcomes assessing participation was investigated over three 28-d study periods: pre-ICU admission and at 3 and 6 months following discharge. The following were calculated: time spent at home; the number of destinations visited; linear distance travelled; and two 'activity spaces', a minimum convex polygon and standard deviation ellipse. RESULTS: Results are median [interquartile range] or n (%). The number of successful extractions was 9/50 (18%), 12/39 (31%), and 13/33 (39%); the percentage of time spent at home was 61 [56-68]%, 77 [66-87]%, and 67 [58-77]% (P = 0.16); the number of destinations visited was 34 [18-64], 38 [22-63], and 65 [46-88] (P = 0.02); linear distance travelled was 367 [56-788], 251 [114-323], and 747 [326-933] km over 28 d (P = 0.02), pre-ICU admission and at 3 and 6 months following ICU discharge, respectively. Activity spaces were successfully created. CONCLUSION: Limited smartphone ownership, missing data, and time-consuming data extraction limit current implementation of mass extraction of location data from patients' smartphones to aid prognostication or measure outcomes. The number of journeys taken and the linear distance travelled increased between 3 and 6 months, suggesting participation may improve over time.

Mixed-data deep learning in repeated predictions of general medicine length of stay: a derivation study.

The accurate prediction of likely discharges and estimates of length of stay (LOS) aid in effective hospital administration and help to prevent access block. Machine learning (ML) may be able to help with these tasks. For consecutive patients admitted under General Medicine at the Royal Adelaide Hospital over an 8-month period, daily ward round notes and relevant discrete data fields were collected from the electronic medical record. These data were then split into training and testing sets (7-month/1-month train/test split) prior to use in ML analyses aiming to predict discharge within the next 2 days, discharge within the next 7 days and an estimated date of discharge (EDD). Artificial neural networks and logistic regression were effective at predicting discharge within 48 h of a given ward round note. These models achieved an area under the receiver operator curve (AUC) of 0.80 and 0.78, respectively. Prediction of discharge within 7 days of a given note was less accurate, with artificial neural network returning an AUC of 0.68 and logistic regression an AUC of 0.61. The generation of an exact EDD remains inaccurate. This study has shown that repeated estimates of LOS using daily ward round notes and mixed-data inputs are effective in the prediction of general medicine discharges in the next 48 h. Further research may seek to prospectively and externally validate models for prediction of upcoming discharge, as well as combination human-ML approaches for generating EDDs.

Evaluation of coagulation status using viscoelastic testing in intensive care patients with coronavirus disease 2019 (COVID-19): An observational point prevalence cohort study.

BACKGROUND: Coronavirus Disease-19 (COVID-19) is associated with a high rate of thrombosis, the pathophysiology of which is not well defined. Viscoelastic testing may identify and characterise hypercoagulable states which are not apparent using conventional coagulation assays. OBJECTIVES: The objective of this study was to undertake viscoelastic evaluation of the coagulation state in critically ill adults with COVID-19-associated respiratory failure METHODS: This was a single-centre observational point prevalence cohort study of adults with COVID-19-associated respiratory failure requiring respiratory support in the intensive care unit. Coagulation status was evaluated using rotational thromboelastometry (ROTEM®) in conjunction with laboratory markers of coagulation. RESULTS: Six patients fulfilled inclusion criteria. Each patient had one ROTEM® performed. All patients had supranormal clot amplitude at 10 min (A10) and supranormal clot firmness (maximal clot firmness) measured in at least one ROTEM® pathway, and five were supranormal on all pathways. Minimal clot lysis was present on all analyses. Fibrinogen and D-dimer were elevated and routine markers of coagulation within normal ranges in all patients. CONCLUSION: Patients with COVID-19-associated respiratory failure admitted to the intensive care unit exhibit a hypercoagulable state which is not appreciable on conventional tests of coagulation. Supranormal clot firmness, minimal fibrinolysis, and hyperfibrinogenaemia are key findings. Further research is required into the pathophysiology of this hypercoagulable state, as well as the harms and benefits of different anticoagulation strategies.

Prediction of general medical admission length of stay with natural language processing and deep learning: a pilot study.

Length of stay (LOS) and discharge destination predictions are key parts of the discharge planning process for general medical hospital inpatients. It is possible that machine learning, using natural language processing, may be able to assist with accurate LOS and discharge destination prediction for this patient group. Emergency department triage and doctor notes were retrospectively collected on consecutive general medical and acute medical unit admissions to a single tertiary hospital from a 2-month period in 2019. These data were used to assess the feasibility of predicting LOS and discharge destination using natural language processing and a variety of machine learning models. 313 patients were included in the study. The artificial neural network achieved the highest accuracy on the primary outcome of predicting whether a patient would remain in hospital for > 2 days (accuracy 0.82, area under the received operator curve 0.75, sensitivity 0.47 and specificity 0.97). When predicting LOS as an exact number of days, the artificial neural network achieved a mean absolute error of 2.9 and a mean squared error of 16.8 on the test set. For the prediction of home as a discharge destination (vs any non-home alternative), all models performed similarly with an accuracy of approximately 0.74. This study supports the feasibility of using natural language processing to predict general medical inpatient LOS and discharge destination. Further research is indicated with larger, more detailed, datasets from multiple centres to optimise and examine the accuracy that may be achieved with such predictions.

An observational study investigating the use of patient-owned technology to quantify physical activity in survivors of critical illness.

BACKGROUND: Physical activity after intensive care unit (ICU) discharge is challenging to measure but could inform research and practice. A patient's smartphone may provide a novel method to quantify physical activity. OBJECTIVES: We aimed to evaluate the feasibility and accuracy of using smartphone step counts among survivors of critical illness. METHODS: We performed a prospective observational cohort study in 50 patients who had an ICU length of stay>48 h, owned a smartphone, were ambulatory before admission, and were likely to attend follow-up at 3 and 6 months after discharge. At follow-up, daily step counts were extracted from participants' smartphones and two FitBit pedometers, and exercise capacity (6-min walk test) and quality of life (European Quality of Life-5 Dimensions) were measured. RESULTS: Thirty-nine (78%) patients returned at 3 months and 33 (66%) at 6 months, the median [interquartile range] smartphone step counts being 3372 [1688-5899] and 2716 [1717-5994], respectively. There was a strong linear relationship, with smartphone approximating 0.71 (0.58, 0.84) of FitBit step counts, P < 0.0001, R-squared = 0.87. There were weak relationships between step counts and the 6-min walk test distance. CONCLUSION: Although smartphone ownership and data acquisition limit the viability of using extracted smartphone steps at this time, mean daily step counts recorded using a smartphone may act as a surrogate for a dedicated pedometer; however, the relationship between step counts and other measures of physical recovery remains unclear.