Readings for 12/5/18
Apologies for the late post! Please put your comments for this week's reading below.
Readings for 9/26/18
Readings for next week are now posted. Although the readings are not ordered, the order listed is probably the best one. Make sure your respond with your comments by 9/25 at noon. A few " goodies " follow: a humorous cartoon about curve fitting, and an example of boosting. The following example of boosting is taken from here . I've removed the code and some of the math below to make it easier to follow. 1. Start with some data. We use the normal distribution. 2. Fit an initial model. Here we fit a line. You could view this as an intercept-only line or a line with a slope of zero. In either case, it's not hard to see that the residuals are balanced above and below the line, which is what line fitting is supposed to do. 3. Fit a model on the residuals from step 2 (step 2 residuals). This time we use a decision stump , which is a decision tree with one branch point. This is the stump we fit: As you can see, any x (the horizontal axis) less than
Brenas et al. (2017)
ReplyDeleteThis article focused on malaria surveillance in Africa and the challenges this surveillance takes on. The authors posit to “present [their] efforts toward the design and development of the semantic interoperability and evolution for malaria analytics platform, with the goal of improving data and semantic interoperability and evolution for malaria surveillance and to support the integration of data across multiple scales” (p. 21605). The authors discuss other state-of-the-art malaria surveillance systems and data sources before introducing their own SIEMA architecture. The SIEMA architecture consists of four tiers including presentation, query, service, and data access. The authors use a series of examples to illustrate how SIEMA functions. Much of the descriptions were a bit complex to me. I think this was more due to many new terms I was not familiar with. In their case study, they analyzed standalone changes first but I was a little concerned with how realistic standalone changes really are in real data. The next section addressed some of my concerns in regard to the description and understanding of the complexity of the ever-changing malaria data. SIEMA, as a platform that can recognize and facilitate changes to enhance interoperability between data sources seems like an important tool in this area of research; although, I think that I am a little confused on how it is actually working. Finally, the authors did address one of my earlier concerns with the mention of future work looking at “developing algorithms to express more complex types of changes as well as concurrent changes in the system” (p. 21617). I would be interested to know how this turns out and what they do to make it be able to handle more complexities.
Shaban-Nejad et al. (2016)
This article focused on the need for “systems that monitor for cases of HAIs” (p. 1) in healthcare systems to improve the health of the patient population as HAIs are common and can be deadly. The platform created by the authors allows this type of monitoring. On page 2, the authors discuss how monitoring can be extensive, costly, and time consuming but could “in theory be automated” (p. 2). Thus, the inspiration for their work. The authors discuss the idea of behavioral economics and the term “nudges” which was mentioned in last week’s article. This idea intrigues me and it is interesting to me how they transferred this idea across disciplines to apply to the idea of nudging healthcare professionals in order to encourage more effective monitoring of HAIs. The HAIKU framework seems to be a viable option for healthcare nudging.
Brenas et al. (2017):
ReplyDeleteThis article sparked so much interest to me. The complexity of the models that were discussed and explained kind of overwhelmed me. I think the density of the data worked with in this project reiterated the need for machine learning and AI (just in comparing the quickness and efficiency of a machine running these algorithms compared to a human, it’s a huge difference). The concept of querying was super interesting to me, especially across the different domains of semantics and languages. I know this was a minuscule aspect of the paper and framework of the document as a whole, but I think the potential behind the idea could have a lot of big implications in research. I wonder if the services of some of these services/models/systems reach beyond the research of malaria analytics? It seems a lot of depth of construction and utilization has occurred but can it be extended to other fields of research? or even similar domains beyond malaria?
Shaban-Nejad et al. (2016):
I am not well versed or familiar with behavioral economics, but I found this article to be very informative for someone who has little to no base knowledge on the topic. Following our conversation last week on nudging, and its contribution to this platform, I found it to be little easier to conceptualize. As a whole, I think the work presented is really interesting and seems very beneficial to the health care program. If researchers are able to implement this framework within other infections and complications, like their end goal seems to be, I think it would prove a strong advancement in healthcare research and analysis.
The Brenas paper tried a research strategy to fight malaria that they called the SIEMA framework. I liked how this idea made data about malaria to control its spread. If the researchers can keep up to date on the changes and trends in malaria then this will aid efforts to treat and prevent it. The SIEMA framework they introduced seemed very complex and hard for me to understand. What is graph transformation?
ReplyDeleteThe Shaban-Nejad paper talk about improving patient care with hospital infections using HAIO. Reading about positive reinforcement using “nudging” reminded me of big nudging in last week’s paper. Trying to predict who gets an infection using known risk factors seems very reasonable and positive to me. An important question is how easily can this model be used by other hospitals around the US and the world.
Brenas 2017
ReplyDeleteI find that the sheer amount of unstructured, chaotic data in the current era of big data can certainly be a problem a great deal of the time. This article proposes a framework that makes malaria-related data significantly easier to obtain and query. With how scattered and differently structured the data on malaria seems to be from their descriptions, this framework seems like it is much needed in the field. I see they used SPARQL in the query tier, but I have never heard of SPARQL. I am still a bit confused as to how it works. Is it relational like SQL or is it closer to NoSQL? It seems like it uses both tabular and graphical type representations, but this was still a bit unclear to me.
Shaban-Nejad 2016
This article discusses another semantic web framework, this time for prediction of Healthcare-Associated Infections. In both this and the Brenas articles, they talk about semantic technologies and a semantic web framework, but I’ve actually never heard of semantic technology at all. What exactly makes them different from information technologies and how are they usually implemented on a practical level? What kinds of representations are required and how are they more dynamic and adaptable? Additionally, I was unfamiliar with the term ontology as used in this way (a non-philosophical way). How is the ontology in this kind of framework actually represented?
Brenas et al. 2017
ReplyDeleteThis article outlined a very complicated method for combining information across databases/data sources and monitoring them for updates. It is interesting to see all the components that must be maintained and connected in order to provide up-to-date information and reminds me of my brief introduction to SQL and PERL through SAS and the nightmares of dealing with large databases and hospital/healthcare records, in particular. I wonder why all this effort would be put into connecting these databases rather than focusing on creating one universal application? For example, while there are language differences and updates to definitions, wouldn’t it be possible to program them into some uniform e.g., web service that, with the aid of translators, could be programmed in the appropriate language? I am POSITIVE I am missing some nuance in the problem, as I’m sure many others have had similar ideas in mind and likely met with financial, technical, or political limitations.
Shaban-Nejad et al. (2016)
This was another interesting application of data management procedures to identify certain events—in this case, the likelihood of healthcare associated infections. Here, I wonder how the post-hoc nature of this influences the efficacy of the program. For example, some of the predictive variables they are looking at include SSI-related terms. If these infections are already seen and noted, how does this service work? It seemed like the goal was to quickly notify individuals of potential infection and, initially, I thought this was done in the service of predicting issues and preventing them. Instead, it seems that the program just reiterates that if an SSI is found, a “nudge” should be sent. Is this not already done when these problems are encountered in healthcare? Again, I am unaware of the operations of the healthcare system and am probably overlooking some nuanced detail, so my questions are probably not that useful.
In the Brenas paper, I’m particularly interested in the graph transformations section. It seems like this the area that is the most adaptable to changes in understanding of malaria communicability, as well as the point at which SIEMA could be adapted to consider any other type of complexly proliferating agent. I’m curious about this adaptability, though. Is it determined by a trial-and-error procedure of prediction accuracy? When nodes are added, deleted, or altered, is it the decision of experts in the field, or is there a formulaic optimization according to the system? Or both?
ReplyDeleteAs for the Shaban-Nejad paper, while I understand that this work is primarily concerned with developing and bridging knowledge engineering and the axiomization of health care, specifically related to HAI’s, but I was curious about how exactly the effectiveness of this system is supposed to be accounted for. The developed system seems to be great for determining people who are at high risk levels and who should be nudged, but it isn’t clear to me that automating the nudge decision process is going to reduce these people’s risk beyond what already takes place in health care. Could it be possible to integrate some sort of feedback loop in which the people who are nudge are tracked as such, and then their data afterwards can be used to assess whether or not they continue to have elevated risk of HAI’s or go on to have elevated risk for any other related illnesses?
Brenas et al. (2017):
ReplyDeleteThe aim of SIEMA is to integrate the multitude of disparate malaria monitoring systems and databases worldwide. Like in the other reading, this integrates the OWL Web Ontology Language. I’m wondering what exactly an “ontology” is (what I do know is that it’s a branch of philosophy, but that meaning of the word is obviously completely different :) ). I have the basic idea that it is a set of representations of entities in some domain. It seems there is an ontology, and there is a separate syntax (SPARQL, in this case). There are also different agents (CCA, CMA, and SMA). Agent also has a very special meaning here, but I’m not precisely sure what it is in this context. How are all these elements defined in this system and how are they interacting? (Obviously my problem is not knowing the assumed background knowledge required to read this text.)
Shaban-Nejad et al. (2016):
The authors build a HAI ontology to facilitate the detection of HAIs in hospital settings. I may be way off here, but—this is an expert system, right? Perhaps it is not a pure expert system, because in addition to semantic rules, it uses statistical analysis. One of the more interesting uses they put statistical analysis to is the revision of the domain knowledge—something that would usually be done by querying human experts and expert texts. I bring this up because we haven’t talked much about expert systems this semester. We’ve mostly been focused on machine learning. How do expert systems work, and in what ways does the one on display here differ by being hybridized with statistical or machine learning techniques? (Unless I’m completely off and this is not an expert system…?)
Shaban-Nejad et al. (2016)
ReplyDeleteThis paper defined a semantic infrastructure to assist healthcare professionals and infection control practitioners to effectively monitor health care associated infections. I think this is a meaningful work. I don't quite understand the density of predicted probability. How do they know whether the model have weak or strong predictive power? And, the authors said that many of the steps in the surveillance of HAI could, in theory, be automated. How would it be?
Brenas et al. 2017
I think this paper goes further than Shaban's paper. They considered the changes of data sources, domain ontologies, database schemas, using Change Capture Agent and Consistency Management Agent to detect changes. And, they cope with the changes using the Service Management Agent. In the figure 1, there was an expert doing some work. I am confused how the expert works in the SIEMA Infrastructure.
Brenas et al. (2017):
ReplyDeleteI found this article very interesting, though I am not sure I fully understand exactly how SIEMA works. I appreciate that the article touched on wanting to expand to be able to work with the multiple languages spoken in the region they are primarily looking at, as I think that these potential language barriers are an important issue, especially when dealing with subjects like tracking illnesses. I also wonder how or if this could be applied outside of their research with malaria, whether that be with other illnesses like whooping cough or measles caused by the growing anti-vax movement, or even going outside of the medical field all together.
Shaban-Nejad et al. (2016):
This article covers what is in my opinion a pretty important topic, as healthcare provider error and HAIs can definitely be an issue in our healthcare system. While obviously some of their criteria were less subjective (i.e. blood tests), I wonder if issues with accuracy in patient self reporting of things like pain or discomfort could affect this, as some people may be able to tolerate more pain than others, which could potentially lead to not catching a sign of infection. Also, while overall I can see identifying at-risk patients as a good thing for preventing or quickly treating HAIs, I would hope that this identifier of being "at risk" would not affect the patient's health insurance, as battles about what is covered and how pre-existing conditions can affect cost have been issues with our system in the past.