Preoperative hemoglobin thresholds for increased risk of "medical" complications in fast-track total hip and knee arthroplasty, a secondary analysis of a machine-learning algorithm.

BACKGROUND: There is increasing evidence that gender-specific hemoglobin thresholds may not be ideal in the surgical population. Thus, preoperative anemia defined as a hemoglobin of <13.0 g/dL is a well-established risk factor in elective surgery. However, few studies have investigated the specific influence of preoperative hemoglobin within a machine-learning model using data from an optimized fast-track surgical setup. STUDY DESIGN AND METHODS: A secondary analysis on the specific influence of preoperative hemoglobin level on a machine-learning model developed for identifying patients at increased risk of a length of stay (LOS) of >4 day or readmissions due to medical complications in fast-track total hip and knee arthroplasty within a well-defined fast-track protocol. To evaluate the effect of hemoglobin on the model we calculated SHaply Additive Explanation (SHAP) values for the 3913 patients from our previous test-dataset and stratified by gender and total hip and knee arthroplasty, respectively. RESULTS: The study period ran from January 2017 to August 2017. Median LOS was 1 day and mean preoperative Hb was 15.5 g/dL (SD:1.5), lower in women (14.9 vs. 16.2 g/dL) and with 30.5% of women versus 12.0% of men having a Hb of <13.0 g/dL. There was a steep increase in SHAP value with a preoperative Hb < 14.8 g/dL, and irrespective of gender age and procedure type. DISCUSSION: A machine-learning model found a hemoglobin threshold of <14.8 g/dL for increased risk of impaired recovery, regardless of gender or age, supporting reevaluation of preoperative anemia thresholds in the elective surgical setting.

Machine-learning vs. logistic regression for preoperative prediction of medical morbidity after fast-track hip and knee arthroplasty-a comparative study.

BACKGROUND: Machine-learning models may improve prediction of length of stay (LOS) and morbidity after surgery. However, few studies include fast-track programs, and most rely on administrative coding with limited follow-up and information on perioperative care. This study investigates potential benefits of a machine-learning model for prediction of postoperative morbidity in fast-track total hip (THA) and knee arthroplasty (TKA). METHODS: Cohort study in consecutive unselected primary THA/TKA between 2014-2017 from seven Danish centers with established fast-track protocols. Preoperative comorbidity and prescribed medication were recorded prospectively and information on length of stay and readmissions was obtained through the Danish National Patient Registry and medical records. We used a machine-learning model (Boosted Decision Trees) based on boosted decision trees with 33 preoperative variables for predicting "medical" morbidity leading to LOS > 4 days or 90-days readmissions and compared to a logistical regression model based on the same variables. We also evaluated two parsimonious models, using the ten most important variables in the full machine-learning and logistic regression models. Data collected between 2014-2016 (n:18,013) was used for model training and data from 2017 (n:3913) was used for testing. Model performances were analyzed using precision, area under receiver operating (AUROC) and precision recall curves (AUPRC), as well as the Mathews Correlation Coefficient. Variable importance was analyzed using Shapley Additive Explanations values. RESULTS: Using a threshold of 20% "risk-patients" (n:782), precision, AUROC and AUPRC were 13.6%, 76.3% and 15.5% vs. 12.4%, 74.7% and 15.6% for the machine-learning and logistic regression model, respectively. The parsimonious machine-learning model performed better than the full logistic regression model. Of the top ten variables, eight were shared between the machine-learning and logistic regression models, but with a considerable age-related variation in importance of specific types of medication. CONCLUSION: A machine-learning model using preoperative characteristics and prescriptions slightly improved identification of patients in high-risk of "medical" complications after fast-track THA and TKA compared to a logistic regression model. Such algorithms could help find a manageable population of patients who may benefit most from intensified perioperative care.

A 2-million-year-old ecosystem in Greenland uncovered by environmental DNA.

Late Pliocene and Early Pleistocene epochs 3.6 to 0.8 million years ago1 had climates resembling those forecasted under future warming2. Palaeoclimatic records show strong polar amplification with mean annual temperatures of 11-19 °C above contemporary values3,4. The biological communities inhabiting the Arctic during this time remain poorly known because fossils are rare5. Here we report an ancient environmental DNA6 (eDNA) record describing the rich plant and animal assemblages of the Kap København Formation in North Greenland, dated to around two million years ago. The record shows an open boreal forest ecosystem with mixed vegetation of poplar, birch and thuja trees, as well as a variety of Arctic and boreal shrubs and herbs, many of which had not previously been detected at the site from macrofossil and pollen records. The DNA record confirms the presence of hare and mitochondrial DNA from animals including mastodons, reindeer, rodents and geese, all ancestral to their present-day and late Pleistocene relatives. The presence of marine species including horseshoe crab and green algae support a warmer climate than today. The reconstructed ecosystem has no modern analogue. The survival of such ancient eDNA probably relates to its binding to mineral surfaces. Our findings open new areas of genetic research, demonstrating that it is possible to track the ecology and evolution of biological communities from two million years ago using ancient eDNA.

Spatial heterogeneity affects predictions from early-curve fitting of pandemic outbreaks: a case study using population data from Denmark.

The modelling of pandemics has become a critical aspect in modern society. Even though artificial intelligence can help the forecast, the implementation of ordinary differential equations which estimate the time development in the number of susceptible, (exposed), infected and recovered (SIR/SEIR) individuals is still important in order to understand the stage of the pandemic. These models are based on simplified assumptions which constitute approximations, but to what extent this are erroneous is not understood since many factors can affect the development. In this paper, we introduce an agent-based model including spatial clustering and heterogeneities in connectivity and infection strength. Based on Danish population data, we estimate how this impacts the early prediction of a pandemic and compare this to the long-term development. Our results show that early phase SEIR model predictions overestimate the peak number of infected and the equilibrium level by at least a factor of two. These results are robust to variations of parameters influencing connection distances and independent of the distribution of infection rates.