Machine learning-based delta check method for detecting misidentification errors in tumor marker tests.

OBJECTIVES: Misidentification errors in tumor marker tests can lead to serious diagnostic and treatment errors. This study aims to develop a method for detecting these errors using a machine learning (ML)-based delta check approach, overcoming limitations of conventional methods. METHODS: We analyzed five tumor marker test results: alpha-fetoprotein (AFP), cancer antigen 19-9 (CA19-9), cancer antigen 125 (CA125), carcinoembryonic antigen (CEA), and prostate-specific antigen (PSA). A total of 246,261 records were used in the analysis. Of these, 179,929 records were used for model training and 66,332 records for performance evaluation. We developed a misidentification error detection model based on the random forest (RF) and deep neural network (DNN) methods. We performed an in silico simulation with 1 % random sample shuffling. The performance of the developed models was evaluated and compared to conventional delta check methods such as delta percent change (DPC), absolute DPC (absDPC), and reference change values (RCV). RESULTS: The DNN model outperformed the RF, DPC, absDPC, and RCV methods in detecting sample misidentification errors. It achieved balanced accuracies of 0.828, 0.842, 0.792, 0.818, and 0.833 for AFP, CA19-9, CA125, CEA, and PSA, respectively. Although the RF method performed better than DPC and absDPC, it showed similar or lower performance compared to RCV. CONCLUSIONS: Our research results demonstrate that an ML-based delta check method can more effectively detect sample misidentification errors compared to conventional delta check methods. In particular, the DNN model demonstrated superior and stable detection performance compared to the RF, DPC, absDPC, and RCV methods.

Is Ultra-Short-Term Heart Rate Variability Valid in Non-static Conditions?

In mobile healthcare, heart rate variability (HRV) is increasingly being used in dynamic patient states. In this situation, shortening of the measurement time is required. This study aimed to validate ultra-short-term HRV in non-static conditions. We conducted electrocardiogram (ECG) measurements at rest, during exercise, and in the post-exercise recovery period in 30 subjects and analyzed ultra-short-term HRV in time and frequency domains by ECG in 10, 30, 60, 120, 180, and 240-s intervals, and compared the values to the 5-min HRV. For statistical analysis, null hypothesis testing, Cohen's d statistics, Pearson's correlation coefficient, and Bland-Altman analysis were used, with a statistical significance level of P < 0.05. The feasibility of ultra-short-term HRV and the minimum time required for analysis showed differences in each condition and for each analysis method. If the strict criteria satisfying all the statistical methods were followed, the ultra-short-term HRV could be derived from a from 30 to 240-s length of ECG. However, at least 120 s was required in the post-exercise recovery or exercise conditions, and even ultra-short-term HRV was not measurable in some variables. In contrast, according to the lenient criteria needed to satisfy only one of the statistical criteria, the minimum time required for ultra-short-term HRV analysis was 10-60 s in the resting condition, 10-180 s in the exercise condition, and 10-120 s in the post-exercise recovery condition. In conclusion, the results of this study showed that a longer measurement time was required for ultra-short-term HRV analysis in dynamic conditions. This suggests that the existing ultra-short-term HRV research results derived from the static condition cannot applied to the non-static conditions of daily life and that a criterion specific to the non-static conditions are necessary.

Photoplethysmogram Analysis and Applications: An Integrative Review.

Beyond its use in a clinical environment, photoplethysmogram (PPG) is increasingly used for measuring the physiological state of an individual in daily life. This review aims to examine existing research on photoplethysmogram concerning its generation mechanisms, measurement principles, clinical applications, noise definition, pre-processing techniques, feature detection techniques, and post-processing techniques for photoplethysmogram processing, especially from an engineering point of view. We performed an extensive search with the PubMed, Google Scholar, Institute of Electrical and Electronics Engineers (IEEE), ScienceDirect, and Web of Science databases. Exclusion conditions did not include the year of publication, but articles not published in English were excluded. Based on 118 articles, we identified four main topics of enabling PPG: (A) PPG waveform, (B) PPG features and clinical applications including basic features based on the original PPG waveform, combined features of PPG, and derivative features of PPG, (C) PPG noise including motion artifact baseline wandering and hypoperfusion, and (D) PPG signal processing including PPG preprocessing, PPG peak detection, and signal quality index. The application field of photoplethysmogram has been extending from the clinical to the mobile environment. Although there is no standardized pre-processing pipeline for PPG signal processing, as PPG data are acquired and accumulated in various ways, the recently proposed machine learning-based method is expected to offer a promising solution.

Postoperative Pain Assessment Model Based on Pulse Contour Characteristics Analysis.

This study aims to develop a new postoperative pain assessment model based on pulse contour analysis and to evaluate its effectiveness in postoperative pain assessment. We derived candidate features from photoplethysmography (PPG) and developed an assessment model based on multiple logistic regressions with a combination of features. This study also includes investigations into the optimal unit of analysis and number of features. For model development, PPGs obtained from 78 surgical patients with a six-min duration in pre- and post-operation conditions, including a training set of 56 pairs and a test set of 22 pairs, were used. We tested models with 5, 10, 20, 30, 40, 50, 60, 70, and 80 beats as an analysis unit, and with 1 to 8 of features for optimization, then determined 20 beats and three features to be the simplest optimal unit of analysis and number of features, respectively. The selected features were RMSSD-ACVonset/ACAbl, AV-Asys/Atotal, and SD-RS, where RMSSD-ACVonset/ACAbl is the root mean square of the successive difference of the ratio of pulse onset amplitude to the pulse onset-to-peak amplitude, AV-Asys/Atotal is the average value of a normalized systolic area of a pulse with a total pulse area, and SD-RS is the standard deviation of a rising slope of a pulse. The accuracy (AC) and the area under the curve (AUC) of the proposed model were 0.793 and 0.872 in the development set (N = 56), respectively, which were superior to those of SPI (AC: 0.643, AUC: 0.716) and ANI (AC: 0.633 AUC: 0.671). In the test set (N = 22), the AC and AUC of the proposed model were 0.712 and 0.808, respectively, which were superior to those of SPI (AC: 0.640, AUC: 0.709) and ANI (AC: 0.640, AUC: 0.680).

Postoperative Pain Assessment Indices Based on Photoplethysmography Waveform Analysis.

The purpose of this study was to derive parameters that might reflect postoperative pain from photoplethysmography (PPG) and verify the derived parameters in postoperative pain assessment. We obtained preoperative and postoperative PPG and 100-mm visual analog scale (VAS) from 65 surgical patients and extracted a total of 51 PPG morphology-based parameters and their normalized parameters from these PPGs obtained. Pain discrimination performances of these derived parameters were assessed by statistical analyses, including Wilcoxon signed rank test with Bonferroni correction, classification accuracy based on logistic regression, and 4-fold cross validation. After comparing these parameters derived from PPG in pre- and post-operative conditions, statistically significant difference was found in 36 of the 51 parameters. Using logistic classification, dynamic between-pulse parameters such as normalized systolic amplitude variation and normalized diastolic amplitude variation showed better pain classification performance than the static within-pulse parameters. VAS score was 0 in every pre-operation condition, but >60 VAS was observed in the post-operative condition. Systolic peak amplitude variation normalized by PPG AC amplitude showed the best performance in classifying post-operative pain, with accuracy, sensitivity, specificity, and positive predictivity values of 79.5, 74.0, 86.0, and 84.5%, respectively. These results are superior to those of the surgical pleth index (SPI, GE Healthcare, Chicago, IL, United States) at 65.9, 65.9, 66.5, and 66.5%, respectively.