skip to main content

Title: The risk of racial bias while tracking influenza-related content on social media using machine learning
Abstract Objective Machine learning is used to understand and track influenza-related content on social media. Because these systems are used at scale, they have the potential to adversely impact the people they are built to help. In this study, we explore the biases of different machine learning methods for the specific task of detecting influenza-related content. We compare the performance of each model on tweets written in Standard American English (SAE) vs African American English (AAE). Materials and Methods Two influenza-related datasets are used to train 3 text classification models (support vector machine, convolutional neural network, bidirectional long short-term memory) with different feature sets. The datasets match real-world scenarios in which there is a large imbalance between SAE and AAE examples. The number of AAE examples for each class ranges from 2% to 5% in both datasets. We also evaluate each model's performance using a balanced dataset via undersampling. Results We find that all of the tested machine learning methods are biased on both datasets. The difference in false positive rates between SAE and AAE examples ranges from 0.01 to 0.35. The difference in the false negative rates ranges from 0.01 to 0.23. We also find that the neural network more » methods generally has more unfair results than the linear support vector machine on the chosen datasets. Conclusions The models that result in the most unfair predictions may vary from dataset to dataset. Practitioners should be aware of the potential harms related to applying machine learning to health-related social media data. At a minimum, we recommend evaluating fairness along with traditional evaluation metrics. « less
Award ID(s):
Publication Date:
Journal Name:
Journal of the American Medical Informatics Association
Sponsoring Org:
National Science Foundation
More Like this
  1. Obeid, Iyad Selesnick (Ed.)
    Electroencephalography (EEG) is a popular clinical monitoring tool used for diagnosing brain-related disorders such as epilepsy [1]. As monitoring EEGs in a critical-care setting is an expensive and tedious task, there is a great interest in developing real-time EEG monitoring tools to improve patient care quality and efficiency [2]. However, clinicians require automatic seizure detection tools that provide decisions with at least 75% sensitivity and less than 1 false alarm (FA) per 24 hours [3]. Some commercial tools recently claim to reach such performance levels, including the Olympic Brainz Monitor [4] and Persyst 14 [5]. In this abstract, we describemore »our efforts to transform a high-performance offline seizure detection system [3] into a low latency real-time or online seizure detection system. An overview of the system is shown in Figure 1. The main difference between an online versus offline system is that an online system should always be causal and has minimum latency which is often defined by domain experts. The offline system, shown in Figure 2, uses two phases of deep learning models with postprocessing [3]. The channel-based long short term memory (LSTM) model (Phase 1 or P1) processes linear frequency cepstral coefficients (LFCC) [6] features from each EEG channel separately. We use the hypotheses generated by the P1 model and create additional features that carry information about the detected events and their confidence. The P2 model uses these additional features and the LFCC features to learn the temporal and spatial aspects of the EEG signals using a hybrid convolutional neural network (CNN) and LSTM model. Finally, Phase 3 aggregates the results from both P1 and P2 before applying a final postprocessing step. The online system implements Phase 1 by taking advantage of the Linux piping mechanism, multithreading techniques, and multi-core processors. To convert Phase 1 into an online system, we divide the system into five major modules: signal preprocessor, feature extractor, event decoder, postprocessor, and visualizer. The system reads 0.1-second frames from each EEG channel and sends them to the feature extractor and the visualizer. The feature extractor generates LFCC features in real time from the streaming EEG signal. Next, the system computes seizure and background probabilities using a channel-based LSTM model and applies a postprocessor to aggregate the detected events across channels. The system then displays the EEG signal and the decisions simultaneously using a visualization module. The online system uses C++, Python, TensorFlow, and PyQtGraph in its implementation. The online system accepts streamed EEG data sampled at 250 Hz as input. The system begins processing the EEG signal by applying a TCP montage [8]. Depending on the type of the montage, the EEG signal can have either 22 or 20 channels. To enable the online operation, we send 0.1-second (25 samples) length frames from each channel of the streamed EEG signal to the feature extractor and the visualizer. Feature extraction is performed sequentially on each channel. The signal preprocessor writes the sample frames into two streams to facilitate these modules. In the first stream, the feature extractor receives the signals using stdin. In parallel, as a second stream, the visualizer shares a user-defined file with the signal preprocessor. This user-defined file holds raw signal information as a buffer for the visualizer. The signal preprocessor writes into the file while the visualizer reads from it. Reading and writing into the same file poses a challenge. The visualizer can start reading while the signal preprocessor is writing into it. To resolve this issue, we utilize a file locking mechanism in the signal preprocessor and visualizer. Each of the processes temporarily locks the file, performs its operation, releases the lock, and tries to obtain the lock after a waiting period. The file locking mechanism ensures that only one process can access the file by prohibiting other processes from reading or writing while one process is modifying the file [9]. The feature extractor uses circular buffers to save 0.3 seconds or 75 samples from each channel for extracting 0.2-second or 50-sample long center-aligned windows. The module generates 8 absolute LFCC features where the zeroth cepstral coefficient is replaced by a temporal domain energy term. For extracting the rest of the features, three pipelines are used. The differential energy feature is calculated in a 0.9-second absolute feature window with a frame size of 0.1 seconds. The difference between the maximum and minimum temporal energy terms is calculated in this range. Then, the first derivative or the delta features are calculated using another 0.9-second window. Finally, the second derivative or delta-delta features are calculated using a 0.3-second window [6]. The differential energy for the delta-delta features is not included. In total, we extract 26 features from the raw sample windows which add 1.1 seconds of delay to the system. We used the Temple University Hospital Seizure Database (TUSZ) v1.2.1 for developing the online system [10]. The statistics for this dataset are shown in Table 1. A channel-based LSTM model was trained using the features derived from the train set using the online feature extractor module. A window-based normalization technique was applied to those features. In the offline model, we scale features by normalizing using the maximum absolute value of a channel [11] before applying a sliding window approach. Since the online system has access to a limited amount of data, we normalize based on the observed window. The model uses the feature vectors with a frame size of 1 second and a window size of 7 seconds. We evaluated the model using the offline P1 postprocessor to determine the efficacy of the delayed features and the window-based normalization technique. As shown by the results of experiments 1 and 4 in Table 2, these changes give us a comparable performance to the offline model. The online event decoder module utilizes this trained model for computing probabilities for the seizure and background classes. These posteriors are then postprocessed to remove spurious detections. The online postprocessor receives and saves 8 seconds of class posteriors in a buffer for further processing. It applies multiple heuristic filters (e.g., probability threshold) to make an overall decision by combining events across the channels. These filters evaluate the average confidence, the duration of a seizure, and the channels where the seizures were observed. The postprocessor delivers the label and confidence to the visualizer. The visualizer starts to display the signal as soon as it gets access to the signal file, as shown in Figure 1 using the “Signal File” and “Visualizer” blocks. Once the visualizer receives the label and confidence for the latest epoch from the postprocessor, it overlays the decision and color codes that epoch. The visualizer uses red for seizure with the label SEIZ and green for the background class with the label BCKG. Once the streaming finishes, the system saves three files: a signal file in which the sample frames are saved in the order they were streamed, a time segmented event (TSE) file with the overall decisions and confidences, and a hypotheses (HYP) file that saves the label and confidence for each epoch. The user can plot the signal and decisions using the signal and HYP files with only the visualizer by enabling appropriate options. For comparing the performance of different stages of development, we used the test set of TUSZ v1.2.1 database. It contains 1015 EEG records of varying duration. The any-overlap performance [12] of the overall system shown in Figure 2 is 40.29% sensitivity with 5.77 FAs per 24 hours. For comparison, the previous state-of-the-art model developed on this database performed at 30.71% sensitivity with 6.77 FAs per 24 hours [3]. The individual performances of the deep learning phases are as follows: Phase 1’s (P1) performance is 39.46% sensitivity and 11.62 FAs per 24 hours, and Phase 2 detects seizures with 41.16% sensitivity and 11.69 FAs per 24 hours. We trained an LSTM model with the delayed features and the window-based normalization technique for developing the online system. Using the offline decoder and postprocessor, the model performed at 36.23% sensitivity with 9.52 FAs per 24 hours. The trained model was then evaluated with the online modules. The current performance of the overall online system is 45.80% sensitivity with 28.14 FAs per 24 hours. Table 2 summarizes the performances of these systems. The performance of the online system deviates from the offline P1 model because the online postprocessor fails to combine the events as the seizure probability fluctuates during an event. The modules in the online system add a total of 11.1 seconds of delay for processing each second of the data, as shown in Figure 3. In practice, we also count the time for loading the model and starting the visualizer block. When we consider these facts, the system consumes 15 seconds to display the first hypothesis. The system detects seizure onsets with an average latency of 15 seconds. Implementing an automatic seizure detection model in real time is not trivial. We used a variety of techniques such as the file locking mechanism, multithreading, circular buffers, real-time event decoding, and signal-decision plotting to realize the system. A video demonstrating the system is available at: The final conference submission will include a more detailed analysis of the online performance of each module. ACKNOWLEDGMENTS Research reported in this publication was most recently supported by the National Science Foundation Partnership for Innovation award number IIP-1827565 and the Pennsylvania Commonwealth Universal Research Enhancement Program (PA CURE). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the official views of any of these organizations. REFERENCES [1] A. Craik, Y. He, and J. L. Contreras-Vidal, “Deep learning for electroencephalogram (EEG) classification tasks: a review,” J. Neural Eng., vol. 16, no. 3, p. 031001, 2019. [2] A. C. Bridi, T. Q. Louro, and R. C. L. Da Silva, “Clinical Alarms in intensive care: implications of alarm fatigue for the safety of patients,” Rev. Lat. Am. Enfermagem, vol. 22, no. 6, p. 1034, 2014. [3] M. Golmohammadi, V. Shah, I. Obeid, and J. Picone, “Deep Learning Approaches for Automatic Seizure Detection from Scalp Electroencephalograms,” in Signal Processing in Medicine and Biology: Emerging Trends in Research and Applications, 1st ed., I. Obeid, I. Selesnick, and J. Picone, Eds. New York, New York, USA: Springer, 2020, pp. 233–274. [4] “CFM Olympic Brainz Monitor.” [Online]. Available: [Accessed: 17-Jul-2020]. [5] M. L. Scheuer, S. B. Wilson, A. Antony, G. Ghearing, A. Urban, and A. I. Bagic, “Seizure Detection: Interreader Agreement and Detection Algorithm Assessments Using a Large Dataset,” J. Clin. Neurophysiol., 2020. [6] A. Harati, M. Golmohammadi, S. Lopez, I. Obeid, and J. Picone, “Improved EEG Event Classification Using Differential Energy,” in Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium, 2015, pp. 1–4. [7] V. Shah, C. Campbell, I. Obeid, and J. Picone, “Improved Spatio-Temporal Modeling in Automated Seizure Detection using Channel-Dependent Posteriors,” Neurocomputing, 2021. [8] W. Tatum, A. Husain, S. Benbadis, and P. Kaplan, Handbook of EEG Interpretation. New York City, New York, USA: Demos Medical Publishing, 2007. [9] D. P. Bovet and C. Marco, Understanding the Linux Kernel, 3rd ed. O’Reilly Media, Inc., 2005. [10] V. Shah et al., “The Temple University Hospital Seizure Detection Corpus,” Front. Neuroinform., vol. 12, pp. 1–6, 2018. [11] F. Pedregosa et al., “Scikit-learn: Machine Learning in Python,” J. Mach. Learn. Res., vol. 12, pp. 2825–2830, 2011. [12] J. Gotman, D. Flanagan, J. Zhang, and B. Rosenblatt, “Automatic seizure detection in the newborn: Methods and initial evaluation,” Electroencephalogr. Clin. Neurophysiol., vol. 103, no. 3, pp. 356–362, 1997.« less
  2. Introduction The Centers for Disease Control and Prevention (CDC) spend significant time and resources to track influenza vaccination coverage each influenza season using national surveys. Emerging data from social media provide an alternative solution to surveillance at both national and local levels of influenza vaccination coverage in near real time. Objectives This study aimed to characterise and analyse the vaccinated population from temporal, demographical and geographical perspectives using automatic classification of vaccination-related Twitter data. Methods In this cross-sectional study, we continuously collected tweets containing both influenza-related terms and vaccine-related terms covering four consecutive influenza seasons from 2013 to 2017. Wemore »created a machine learning classifier to identify relevant tweets, then evaluated the approach by comparing to data from the CDC’s FluVaxView. We limited our analysis to tweets geolocated within the USA. Results We assessed 1 124 839 tweets. We found strong correlations of 0.799 between monthly Twitter estimates and CDC, with correlations as high as 0.950 in individual influenza seasons. We also found that our approach obtained geographical correlations of 0.387 at the US state level and 0.467 at the regional level. Finally, we found a higher level of influenza vaccine tweets among female users than male users, also consistent with the results of CDC surveys on vaccine uptake. Conclusion Significant correlations between Twitter data and CDC data show the potential of using social media for vaccination surveillance. Temporal variability is captured better than geographical and demographical variability. We discuss potential paths forward for leveraging this approach.« less
  3. Background Internet data can be used to improve infectious disease models. However, the representativeness and individual-level validity of internet-derived measures are largely unexplored as this requires ground truth data for study. Objective This study sought to identify relationships between Web-based behaviors and/or conversation topics and health status using a ground truth, survey-based dataset. Methods This study leveraged a unique dataset of self-reported surveys, microbiological laboratory tests, and social media data from the same individuals toward understanding the validity of individual-level constructs pertaining to influenza-like illness in social media data. Logistic regression models were used to identify illness in Twitter postsmore »using user posting behaviors and topic model features extracted from users’ tweets. Results Of 396 original study participants, only 81 met the inclusion criteria for this study. Of these participants’ tweets, we identified only two instances that were related to health and occurred within 2 weeks (before or after) of a survey indicating symptoms. It was not possible to predict when participants reported symptoms using features derived from topic models (area under the curve [AUC]=0.51; P=.38), though it was possible using behavior features, albeit with a very small effect size (AUC=0.53; P≤.001). Individual symptoms were also generally not predictable either. The study sample and a random sample from Twitter are predictably different on held-out data (AUC=0.67; P≤.001), meaning that the content posted by people who participated in this study was predictably different from that posted by random Twitter users. Individuals in the random sample and the GoViral sample used Twitter with similar frequencies (similar @ mentions, number of tweets, and number of retweets; AUC=0.50; P=.19). Conclusions To our knowledge, this is the first instance of an attempt to use a ground truth dataset to validate infectious disease observations in social media data. The lack of signal, the lack of predictability among behaviors or topics, and the demonstrated volunteer bias in the study population are important findings for the large and growing body of disease surveillance using internet-sourced data.« less
  4. Abstract Expert testimony varies in scientific quality and jurors have a difficult time evaluating evidence quality (McAuliff et al., 2009). In the current study, we apply Fuzzy Trace Theory principles, examining whether visual and gist aids help jurors calibrate to the strength of scientific evidence. Additionally we were interested in the role of jurors’ individual differences in scientific reasoning skills in their understanding of case evidence. Contrary to our preregistered hypotheses, there was no effect of evidence condition or gist aid on evidence understanding. However, individual differences between jurors’ numeracy skills predicted evidence understanding. Summary Poor-quality expert evidence is sometimesmore »admitted into court (Smithburn, 2004). Jurors’ calibration to evidence strength varies widely and is not robustly understood. For instance, previous research has established jurors lack understanding of the role of control groups, confounds, and sample sizes in scientific research (McAuliff, Kovera, & Nunez, 2009; Mill, Gray, & Mandel, 1994). Still others have found that jurors can distinguish weak from strong evidence when the evidence is presented alone, yet not when simultaneously presented with case details (Smith, Bull, & Holliday, 2011). This research highlights the need to present evidence to jurors in a way they can understand. Fuzzy Trace Theory purports that people encode information in exact, verbatim representations and through “gist” representations, which represent summary of meaning (Reyna & Brainerd, 1995). It is possible that the presenting complex scientific evidence to people with verbatim content or appealing to the gist, or bottom-line meaning of the information may influence juror understanding of that evidence. Application of Fuzzy Trace Theory in the medical field has shown that gist representations are beneficial for helping laypeople better understand risk and benefits of medical treatment (Brust-Renck, Reyna, Wilhelms, & Lazar, 2016). Yet, little research has applied Fuzzy Trace Theory to information comprehension and application within the context of a jury (c.f. Reyna et. al., 2015). Additionally, it is likely that jurors’ individual characteristics, such as scientific reasoning abilities and cognitive tendencies, influence their ability to understand and apply complex scientific information (Coutinho, 2006). Methods The purpose of this study was to examine how jurors calibrate to the strength of scientific information, and whether individual difference variables and gist aids inspired by Fuzzy Trace Theory help jurors better understand complicated science of differing quality. We used a 2 (quality of scientific evidence: high vs. low) x 2 (decision aid to improve calibration - gist information vs. no gist information), between-subjects design. All hypotheses were preregistered on the Open Science Framework. Jury-eligible community participants (430 jurors across 90 juries; Mage = 37.58, SD = 16.17, 58% female, 56.93% White). Each jury was randomly assigned to one of the four possible conditions. Participants were asked to individually fill out measures related to their scientific reasoning skills prior to watching a mock jury trial. The trial was about an armed bank robbery and consisted of various pieces of testimony and evidence (e.g. an eyewitness testimony, police lineup identification, and a sweatshirt found with the stolen bank money). The key piece of evidence was mitochondrial DNA (mtDNA) evidence collected from hair on a sweatshirt (materials from Hans et al., 2011). Two experts presented opposing opinions about the scientific evidence related to the mtDNA match estimate for the defendant’s identification. The quality and content of this mtDNA evidence differed based on the two conditions. The high quality evidence condition used a larger database than the low quality evidence to compare to the mtDNA sample and could exclude a larger percentage of people. In the decision aid condition, experts in the gist information group presented gist aid inspired visuals and examples to help explain the proportion of people that could not be excluded as a match. Those in the no gist information group were not given any aid to help them understand the mtDNA evidence presented. After viewing the trial, participants filled out a questionnaire on how well they understood the mtDNA evidence and their overall judgments of the case (e.g. verdict, witness credibility, scientific evidence strength). They filled this questionnaire out again after a 45-minute deliberation. Measures We measured Attitudes Toward Science (ATS) with indices of scientific promise and scientific reservations (Hans et al., 2011; originally developed by National Science Board, 2004; 2006). We used Drummond and Fischhoff’s (2015) Scientific Reasoning Scale (SRS) to measure scientific reasoning skills. Weller et al.’s (2012) Numeracy Scale (WNS) measured proficiency in reasoning with quantitative information. The NFC-Short Form (Cacioppo et al., 1984) measured need for cognition. We developed a 20-item multiple-choice comprehension test for the mtDNA scientific information in the cases (modeled on Hans et al., 2011, and McAuliff et al., 2009). Participants were shown 20 statements related to DNA evidence and asked whether these statements were True or False. The test was then scored out of 20 points. Results For this project, we measured calibration to the scientific evidence in a few different ways. We are building a full model with these various operationalizations to be presented at APLS, but focus only on one of the calibration DVs (i.e., objective understanding of the mtDNA evidence) in the current proposal. We conducted a general linear model with total score on the mtDNA understanding measure as the DV and quality of scientific evidence condition, decision aid condition, and the four individual difference measures (i.e., NFC, ATS, WNS, and SRS) as predictors. Contrary to our main hypotheses, neither evidence quality nor decision aid condition affected juror understanding. However, the individual difference variables did: we found significant main effects for Scientific Reasoning Skills, F(1, 427) = 16.03, p <.001, np2 = .04, Weller Numeracy Scale, F(1, 427) = 15.19, p <.001, np2 = .03, and Need for Cognition, F(1, 427) = 16.80, p <.001, np2 = .04, such that those who scored higher on these measures displayed better understanding of the scientific evidence. In addition there was a significant interaction of evidence quality condition and scores on the Weller’s Numeracy Scale, F(1, 427) = 4.10, p = .04, np2 = .01. Further results will be discussed. Discussion These data suggest jurors are not sensitive to differences in the quality of scientific mtDNA evidence, and also that our attempt at helping sensitize them with Fuzzy Trace Theory-inspired aids did not improve calibration. Individual scientific reasoning abilities and general cognition styles were better predictors of understanding this scientific information. These results suggest a need for further exploration of approaches to help jurors differentiate between high and low quality evidence. Note: The 3rd author was supported by an AP-LS AP Award for her role in this research. Learning Objective: Participants will be able to describe how individual differences in scientific reasoning skills help jurors understand complex scientific evidence.« less
  5. Abstract
    Excessive phosphorus (P) applications to croplands can contribute to eutrophication of surface waters through surface runoff and subsurface (leaching) losses. We analyzed leaching losses of total dissolved P (TDP) from no-till corn, hybrid poplar (Populus nigra X P. maximowiczii), switchgrass (Panicum virgatum), miscanthus (Miscanthus giganteus), native grasses, and restored prairie, all planted in 2008 on former cropland in Michigan, USA. All crops except corn (13 kg P ha−1 year−1) were grown without P fertilization. Biomass was harvested at the end of each growing season except for poplar. Soil water at 1.2 m depth was sampled weekly to biweekly for TDP determination during March–November 2009–2016More>>