Optimal Windows for Actionable Glucose Monitoring

A CGM (Continuous Glucose Monitor) sensor monitors and reports glucose every few minutes. Typical sensors (worn under the skin) last about 2 weeks. A Lab based HbA1C test measures a snapshot of this average in a single test. Ideally they tend to be in the ballpark. 
Image source: Gen AI

Quick Recap

Glucose monitoring for clinical evaluation has come a long way since measuring one's 'fasting glucose'. This value provided a snapshot of your resting glucose levels early in the morning after allowing for overnight resumption to baseline levels. 

One also had the Oral Glucose Tolerance Test (OGTT), though this was a more detailed procedure - ~ 2hrs and many pokes in between to sample your blood. A graph plot of glucose levels in this time period showed how the body was responding to a controlled glucose challenge. 

These two tests were good for diagnosis and getting a picture of glucose homeostasis in an individual. Based on this healthcare professionals (HCPs) could decide on course of action to manage cases of glucose intolerance a.k.a diabetes (or even prediabetes).

Enter HbA1C test OR more simply A1C test. This test changed the paradigm of evaluation and management of diabetes in clinical settings. It provided a snapshot average of glucose levels in the body over three months. Useful for long-term glucose management. 

So why was this cool? No more asking a patient to arrive fasting at the lab. Also the levels gave a clear window into how the glucose homeostasis was over the three months. This could assist HCPs in providing more actionable suggestions in terms of managing lifestyle choices in patients e.g. specific foods, physical activities etc.

The principle behind the test was elegant and simple. Hemoglobin in the RBC cells in blood reacted with glucose at a constant rate to form a product that could be easily measured. Since this simple chemical reaction occurred in the body at a constant rate one could calibrate the glucose levels easily. Why 3 months? Remember RBCs in our body have a lifespan of about 120 days. So a test after another 120 days was enough to monitor the progress. 

Well! What if the lifespan of RBCs aren't ever the same? What if the RBCs of certain demographics are inherently different to reacting with glucose? What about cases of hemoglobinopathies?

The answer is yes to all of the above. So, though Hemoglobin A1c (HbA1c) has come to be regarded as the gold standard for assessing glycemic control in individuals with diabetes, it is not without its limitations. Regardless of the caveats the test continues to be useful and a mainstream approach to evaluation of glucose homeostasis in an individual. 

The Problem with Traditional A1C

One significant caveat is the demographic differences observed in HbA1c levels. For instance, research indicates that South Asians tend to have HbA1c levels that are approximately 0.5–0.7% higher than other populations for the same glucose exposure [1]. This discrepancy can lead to misinterpretations of glycemic control and potentially inappropriate clinical decisions.

Moreover, biological confounders such as variations in red cell turnover and the presence of hemoglobinopathies can affect HbA1c readings [2]. Conditions like anemia or sickle cell disease can alter the lifespan of erythrocytes, thereby impacting the accuracy of HbA1c measurements. These factors underscore the need for more reliable and individualized methods of monitoring glucose levels.

Given these limitations, the introduction of Continuous Glucose Monitoring (CGM) systems offers a promising alternative. CGM provides real-time data on glucose levels, capturing fluctuations that HbA1c might miss [3]. This leads us to the next section, where we explore the strengths of CGM data and its potential to revolutionize diabetes management.

Table: Comparison of HbA1c and CGM Metrics

Metric	HbA1c	CGM
Data Points	Average over 2-3 months	~25,920 data points in 90 days
Influencing Factors	Red cell turnover, hemoglobinopathies	Real-time glucose fluctuations
Demographic Differences	Higher A1C in South Asians	Individualized data

2. The Strength of CGM Data: Power of Numbers

Continuous Glucose Monitoring (CGM) systems have transformed diabetes management by providing a wealth of data that far surpasses traditional methods. Over a 90-day period, CGM devices can generate approximately 25,920 data points, offering a detailed and dynamic picture of glucose levels [4]. This abundance of data significantly reduces random error, averaging out the Mean Absolute Relative Difference (MARD) to around 8.3%.

The power of CGM lies in its ability to capture real-time glucose fluctuations, which HbA1c measurements might miss. Unlike HbA1c, which reflects an average over several months, CGM provides immediate insights into daily glucose patterns [5]. This real-time data is invaluable for making timely therapeutic adjustments and for understanding the impact of lifestyle changes on glucose levels.

Moreover, the CGM-derived Glucose Management Indicator (GMI) correlates well with HbA1c but avoids the biases linked to erythrocyte lifespan [6]. This makes GMI a more reliable indicator for individuals with conditions that affect red cell turnover or hemoglobinopathies. By providing a more accurate reflection of glucose control, CGM can help tailor treatment plans more effectively.

As we delve deeper into the capabilities of CGM, it's important to consider the limitations of simply accumulating more data. In the next section, we will explore why extending the monitoring period to six months may not necessarily yield better clinical outcomes and how shorter data windows can be more actionable.

Table: Strengths of CGM vs. HbA1c

Feature	HbA1c	CGM
Data Frequency	Every 2-3 months	Continuous
Error Reduction	Subject to random error	Reduces random error (8.3% MARD)
Correlation	Reflects average glucose	GMI correlates well with A1C

3. The Limits of "More Data"—Why 6 Months May Not Be Better

While the extensive data provided by Continuous Glucose Monitoring (CGM) systems is invaluable, there are diminishing returns when extending the monitoring period beyond 90 days. Although a longer duration might seem beneficial for capturing comprehensive glucose trends, it can actually obscure meaningful outliers and reduce the clinical relevance of the data [7].

One key issue with extending CGM data to 180 days is the potential for older data to dilute the impact of more recent glucose fluctuations. Therapeutic adjustments, such as insulin titration, require timely and accurate data to be effective [8]. Relying on older data can lead to less responsive and potentially suboptimal treatment decisions. For instance, a patient's glucose levels might have stabilized or worsened in the recent past, but this change could be masked by the inclusion of older data.

Behavioral coaching also benefits from shorter data windows. Immediate feedback loops are crucial for encouraging patients to make lifestyle changes and adhere to treatment plans [9]. Shorter monitoring periods, such as 14-30 days, provide actionable insights that can motivate patients and help healthcare providers make more precise adjustments.

However, longer-term trends are not without their value. They are useful for risk stratification and understanding the overall trajectory of a patient's glucose control [10]. Yet, the key is to balance the need for long-term data with the necessity for timely interventions. This leads us to the next section, where we will discuss the optimal data windows for different clinical use cases and how to tailor CGM data interpretation to maximize its utility.

Table: Impact of Data Window Length

Data Window	Benefits	Drawbacks
14-30 days	Immediate feedback, acute adjustments	Limited long-term trends
90 days	Aligns with A1C, comprehensive monitoring	May miss recent changes
180+ days	Long-term trends, risk stratification	Obscures recent data, less responsive

4. Optimal Data Windows for Different Use Cases

Determining the optimal data window for Continuous Glucose Monitoring (CGM) is crucial for maximizing its clinical utility. Different use cases require different monitoring periods to ensure the data is actionable and relevant.

14-30 days: This shorter window is ideal for acute treatment adjustments, such as basal insulin titration. It provides immediate insights into the patient's current glucose levels, allowing for timely and precise therapeutic changes. Short-term data is also beneficial for behavioral coaching, as it offers quick feedback that can motivate patients and help healthcare providers make more precise adjustments.

90 days: Aligning with the traditional HbA1c measurement period, a 90-day window is best for clinical monitoring. It balances the need for comprehensive data with the necessity for timely interventions. This period smooths out random errors and provides a reliable average of glucose levels, making it suitable for routine check-ups and ongoing management of diabetes.

180+ days: Longer-term data windows are useful for observing trends and risk stratification. They help healthcare providers understand the overall trajectory of a patient's glucose control and identify patterns that might indicate future complications. However, while valuable for long-term planning, these extended periods are less effective for immediate therapeutic adjustments and can dull responsiveness to current conditions.

This approach ensures that the data is both relevant and actionable, enhancing the effectiveness of lifestyle adaptions while managing diabetes.

5. Concluding Thoughts

The journey through understanding temporal granularity in glucose monitoring highlights the importance of selecting the optimal data window for clinical actionability. Continuous Glucose Monitoring (CGM) systems offer a powerful alternative to traditional HbA1c measurements, providing real-time data that can significantly enhance diabetes management. However, the key to maximizing the utility of CGM lies in tailoring the data window to the specific clinical needs.

Take Home Messages:

1. Recognize the Limitations of HbA1c: While HbA1c is a valuable tool, it has significant limitations due to demographic differences and biological confounders. CGM offers a more individualized and accurate reflection of glucose control.
2. Leverage the Strength of CGM Data: The extensive data provided by CGM systems reduces random error and offers immediate insights into glucose fluctuations, making it a superior tool for real-time diabetes management.
3. Balance Data Windows for Clinical Relevance: Shorter data windows (14-30 days) are ideal for acute treatment adjustments and behavioral coaching, while 90-day periods align well with traditional monitoring practices. Longer-term data (180+ days) is useful for risk stratification but less effective for immediate interventions.

Reconciling BGM and CGM Accuracy: Can the Twain Meet in Glucose Monitoring?

Have you ever wondered why your Continuous Glucose Monitor (CGM) and Blood Glucose Monitor (BGM) seem to give slightly different readings?

A Continuous Glucose Monitor (CGM) patch worn over the skin 

If you’re living with diabetes or monitoring your blood glucose regularly, you’ve likely experienced this. You may be comparing the numbers from your CGM, a device that continuously tracks glucose levels, to those from a traditional fingerstick BGM, and notice that the values aren’t exactly the same. With their stated accuracy ranges, it may seem that neither can be dependable. Yet the real world readings look more closer. Assuming the provided ranges to be conservative can one work out more realistic ranges? Let's dig in some Math here.

The Problem: Why Do These Devices Disagree?

Both CGMs and BGMs have some level of error in their measurements. Manufacturers often provide accuracy percentages to help users understand this. For example, a CGM might claim to be accurate within 8.5%, while a BGM could have an accuracy rate of 15%. These percentages mean that the devices can be off by that much from the actual blood glucose level.

However, if you assume that one device is reading at the highest possible value of its error range and the other at the lowest, you might think the difference between the two readings could be huge. Yet, in real life, users often report that the numbers are closer than expected—sometimes differing by only 5%. So why is the real-world experience so much better than the theoretical extremes?

Let’s take a closer look with a bit of math.

The Math: Finding the Real Discrepancy

Manufacturers usually advertise the maximum error range, but in practice, these errors don’t always need to occur at their worst levels. Instead, the devices could perform more reliably, leading to smaller discrepancies.

Imagine you’re looking at a CGM with an 8.5% error and a BGM with a 15% error. When both devices are calibrated and used in real-world conditions, these error percentages give us a combined potential error for both devices. The key here is to understand how these errors interact.

I asked my friend who helped me out with the math a bit. Rather than just adding them up, we can use a technique from statistics called root mean square error (RMS). It helps us estimate the combined effect of these errors when they’re not always acting in the worst possible way. The formula is:

E_combined = √{(E_CGM)² + (E_BGM)²}

In other words it is the sum of the squares of the error percentages followed by taking the square root of that sum. Substituting in the values one approximately ends up with approx 17.25%

This means that, in theory, the maximum combined error could be as high as 17.25%. But remember, this is only the worst-case scenario.

Real-World Discrepancies: Why Do They Seem Lower?

In practice, you may notice that your readings from the CGM and BGM differ by far less than this combined error. Instead of seeing huge differences, you might see discrepancies of around 5%, which is a lot tighter than the 17% we just calculated.

To understand why this happens, let’s assume the real-world errors for both devices are smaller than the advertised extremes. If your CGM is performing better than expected, say with an effective error of 2.46%, and your BGM is similarly more accurate with an effective error of 4.34%, the observed discrepancy will shrink.

When these lower error rates are plugged into the same formula, the combined error becomes closer to what you actually observe—around 5%. This reflects the real-world behavior of the devices, where both are more precise during stable glucose periods, reducing the gap between their readings.

The Solution: What Can You Expect?

So, what’s the takeaway? While manufacturers provide maximum error ranges to be cautious, your real-world experience with these devices is usually more consistent. Rather than facing a 17% discrepancy, you’re likely to see much smaller differences, especially during stable glucose levels.

Knowing this can bring some peace of mind when using both a CGM and a BGM. If your readings are close—within 5%—that’s a sign that both devices are functioning well. And if you do see larger discrepancies, it’s a reminder that the devices operate within a known margin of error, and it’s not always cause for concern.

In Conclusion

Understanding the science and math behind blood glucose monitors can help us better interpret the data we see every day. So, whether your CGM and BGM readings are best friends or frenemies, what’s been your experience? Let’s see if your glucose monitors can agree more often than your favorite sitcom characters!

Sometimes, all it takes is a little math to make sense of it all!

Sample image of a Blood Glucose Meter. Image generated using OpenAI's DALL·E tool

Getting started with astrophotography on a budget

 

Ready for the Night skies and wonders it will unfold. Image credits: Self

Quite often I am asked, "How much does it cost to be in this hobby of astrophotography". The questions may vary in their tone and intent but they all finally come down to this particular point - COST. Another related one I often encounter is 'What is the best telescope?' to either get started with OR have one for a long time.

Learning about the the night skies visually is in fact an excellent way to get started. Call it Skywatching, Star hopping, or what you will, it is all about looking at the star constellations, identifying them, and recognizing the names attributed to them in local cultures, conventional astronomy, etc. I started with mine decades ago, when my Dad took me out to show the Saptarishi (Big Dipper), and Orion belt, watch the Grahanams (eclipses), create pin-hole cameras, DIY telescopes with lenses and tubes, explain the night skies, and sharing legends from the Puranas. All that involved $0 cost and lots of bonding time, something more precious than anything else.

For visual observations, a good pair of astronomy binoculars (10x50 to 20x80) or a small beginner telescope can reveal bright Messier objects, planets, and constellations. While a pair of binoculars can range anywhere between ($30 - $100) some decent beginner scopes can come for way less than one may think, somewhere in the range of $100-$200. These can include Newtonian reflectors compact Galilean type refractors or even some tabletop Dobsonians. There are even some Cassegrain reflector models that start in the sub $200 range though they tend to average higher. Once you start adding accessories like smartphone adapters, filters, and eyepieces, very soon you find yourself getting sucked into the black hole of this money-guzzling, yet amazing hobby.

If you ask an amateur hobbyist who has been in this area for a while, you will find that both of these questions are 'loaded', and come with a lot of 'it depends ...' 'well ...' 'but...' etc. Even a good salesperson in this field might probably start the same way. Yet for those wanting a number and a model, without all the crucial caveats, that number would be $5000 to even $10,000. That would include the scope, the tracker mount, and the imaging system a.k.a camera and accessories. 

Not included in this are the countless hours you spend watching all those amazing YouTube video tutorials online and countless groups on Social Media that selflessly share their nuggets of knowledge with you. Be careful sharing this new-found optimism :).Your friends and family might occasionally get a bit bored OR miffed when you start ghosting them at parties or running home errands.

Yes! Astrophotography can seem daunting when you hear such price tags of $5,000, or even more, for a basic starter kit. But don't let these numbers deter you! The hobby can be explored at many budget levels, depending on your goals. 

Getting started isn’t that costly BUT upgrading is!

Yet again, anyone can get started with astrophotography with just a point-and-click camera and a tripod. Even a smartphone has come a long way. Don’t be surprised to find mind-blowing images of the Milky Way OR Andromeda galaxy taken with just a smartphone and a tripod. If you are the one having that DSLR camera you haven’t touched in a while, it’s probably time to dust it off and grab your tripod. Get ready to do a bit more detailed imaging of some brighter objects like the Orion Nebula. Before you realize it, you are already knee-deep into astrophotography, probably even more. Objects like the mineral Moon, our Milky Way, Andromeda Galaxy, and Orion Nebula might soon be a part of your album! The results can sometimes surprise you, pleasantly of course, and will also allow you to learn techniques in image stacking, image pre-processing, and doing the final edits. There are many open-source software that present themselves with varying levels of learning curves. How one can go about doing this is a separate topic. So the big numbers in costs that I mentioned right at the outset can wait, while you can get started with gear that you probably have already..

Image: Image revealing the mineral features of the Moon. A simple DSLR on a tripod is enough to create this type of image. This is my image taken using an AT60mm ED refractor attached to my Canon DSLR camera. Image acquisition, processing, and final editing form the final part of this process. Published in SkyandTelescope

As for ‘which telescope is the best’ or ‘ideal’, such a question calls for a separate discussion that can even extend to a few sessions. However, remember that the main rule is that the cost of the scope itself forms only a part of the entire rig. It would be roughly 1/3rd to 1/4th of your budget. Today many good scopes can range anywhere between $500 to $2000. The same goes for the mounts for telescopes too. Remember that no astrophotography is complete without a good tracking mount. This setup allows you to track your celestial object as it makes its way through the night skies. Mounts can be of different types. Most basic ones are simple and are driven by a motor drive OR advanced ones that slew your set up right to the target and track it accurately. Costs typically range in between $200 - $1000. The latter are called GoTo mounts. Of late there are also some advanced mounts based on Harmonic drives. Such mounts obviate the need for counterweights to a great extent and yet present themselves as very lightweight. The prices of such mounts can go well above $2000. 

Most often it is good to have a couple of scopes. One is typically for DSO imaging (Deep Sky Objects) and the other for Lunar and Planetary imaging. These are mainly classified and recognized by the f-ratios they offer. Lower f-ratios are best suited for DSOs while those with higher (f-ratio > 10-11) are ideal for observing details of Jupiter, Saturn, and Mars. I know of recovering telescope addicts who have a garage full of telescopes, but that calls for a separate light-hearted discussion :D.  Cooling, guiding, and other accessories also add to costs. A Quad-band pass filter alone can cost up to that of a simple beginner telescope. As to why that is needed! You need to check up on something called light pollution of the night skies - a loaded topic that affects our planet's health too. So it's shockingly easy to keep buying accessories as your upgrading of skills progresses.

Manually slewing my rig to the celestial target

While these mid-level scopes are excellent for many targets, some choose to invest $2000-$5000 on semi-pro setups that can reach dim nebulae and galaxies. The latest in the market is all about EAA - Electronically Assisted Astronomy. No Eyepieces! Just automated devices with varying levels of optics. A bit distracting for old-timers in the hobby who also wish to see through the scope with their eyepieces. Even a total novice can get started with just one click and the device starts imaging your desired target, and even takes you on a night sky tour. There are models for every budget ranging from $500 to $5000 and probably more. Astrophotography can become an expensive obsession over time. But by starting small and focusing on skills rather than gear, the universe can be explored at many budget levels. Passion and patience are the most important ingredients.

Gaganam Gaganakaram - For the expanse of the universe can be only compared to itself alone

Good-Bot Bad-Bot: Addressing bias in chatbots and public health information

The following content is available as a preprint on Authorea with the following doi:
https://doi.org/10.22541/au.169030053.36880319/v1 

Abstract

The proliferation of chatbots in recent months has raised concerns about the potential for bias in these conversations. This is especially true when it comes to public health topics, where accurate and unbiased information is essential. Herein, the focus is on the issue of bias in chatbot conversations using oral contraceptive pills (OCPs) as an example. By raising awareness about this issue and emphasizing the need for critical evaluation, we can empower individuals to navigate the digital landscape with confidence and make informed decisions about their health.

Keywords:  

Artificial Intelligence, Bias, Bard, Chatbot, ChatGPT, Health Policy

Introduction

The internet has become the go-to source for health information for many individuals, with search engines like Google being the starting point. However, search results often lead to misleading or inaccurate content. To address this issue, conversational agents like chatbots are emerging as an alternative source of health information. Popular chatbots like ChatGPT and Google’s Bard promise to provide trustworthy and unbiased information on any topic through natural conversations (Fig 1).

Representative image of a user chatting with a virtual assistant guided by an AI-based chatbot. Image source: Bing image creator

However, these chatbots can exhibit biases that shape the health information they provide. Take oral contraceptives (birth control pills) for example. When asked about the pill, chatbots often provide information that skews towards the benefits like pregnancy prevention, while minimizing discussion of potential side effects. This presents a limited perspective on oral contraceptives. The algorithms driving chatbots are trained on available data, which suffers from reporting and publication biases that accentuate benefits over harms. So chatbots end up perpetuating these biases.

Providing comprehensive, balanced information is vital for truly informed decision-making about health. Biased information from chatbots can steer choices in a particular direction, often aligned with business interests rather than public health goals.

To counter such biases, chatbots can employ oversight from experts to ensure balance in the information provided. Guidelines can be issued urging chatbot creators to minimize biases through training data selection and algorithm tweaking. For example, they can be trained to request the user if they need a more comprehensive view in case a request is pointed in one direction.

Finally, educating people to approach chatbots critically rather than blindly trusting their guidance is key. Just like with human experts, examining chatbot recommendations against alternate credible sources allows for balanced perspectives.

In an evolving digital health landscape, chatbots hold promise in improving access to information. But thoughtfully addressing their limitations is crucial so these tools empower rather than inadvertently mislead people in making health choices aligned with their needs. Openness to oversight and continual learning will allow chatbots to better serve individuals and the public health good.

Methods: Understanding inherent bias in popular chatbots

If Internet search engines utilize complex algorithms to deliver their search results then chatbots like ChatGPT and Bard are even more so and at an advanced level. Concepts like Artificial Neural Networks (ANN) and Natural Language Processing (NLP) are just the surface of it. For example, Fig 2 represents a schematic of an affective conversation where the emotion depends on the context. The health assistant understands the affective state of the user in order to generate effective and empathetic responses.

To understand a chat was initiated with ChatGPT, a popular chatbot available at https://chat.openai.com/. At this point, one must understand that all conversations are not the same; hence, responses to the same question posed by other users can evoke different answers. Though this approach personalizes the answers to suit each user and their inherent ‘intent’, the overall objectivity of the answers provided can vary widely. It is entirely possible that responses from AI are in the auto-learning process and keep adapting as the number of users asking the question changes. This makes the process very personalized though not necessarily uniformly objective.

Illustration of an ‘affective’ conversation where the emotion depends on the context. Health assistant understands ‘affective’ state of the user in order to generate ‘affective’ and ‘empathetic’ responses. Image source: Wikipedia. Original source: Ghosal et al.

Results: Exploring slants in a directed inquiry-based conversation

To further illustrate the biases, let us delve deeper into the example of my recent chat with ChatGPT regarding the FDA approval of an oral contraceptive pill containing Progestin. While major media outlets covered this news in a predictable way, curiosity prompted us to investigate the possible health risks associated with hormone-based pills. Recognizing the hormonal nature of such contraceptives, it was reasonable to anticipate the existence of risks and seek comprehensive information on the topic. To be fair though, the approach of using oral contraceptives has been in vogue for decades and has proved helpful in supporting women and their reproductive health.

However, as we embarked on a casual chat, we quickly discovered that the information presented was far from comprehensive or unbiased. The chat grew increasingly in favor of the use of the approved pills while I was seeking information, in particular, about the risks associated with its use. Even pointed requests to provide links from Pubmed failed to give satisfactory results. When we countered it by providing it a copy of an abstract text from a very good recent review paper (that itself was a result of many meta-analyses and papers), it evaluated the paper well while still being defensive about what it said earlier. Finally, it had no choice but to accept there are different sides to the issue as well.

The societal implications, politics, and ensuing interests surrounding birth control contribute to an inherent imbalance in the available literature. This imbalance can result in a lack of representation of all sides of the issue, hindering individuals’ ability to access a diverse range of viewpoints and evidence.

In this particular case, the push to promote the use of oral contraceptive pills, driven by factors such as gender equality, reproductive rights, and public health initiatives, can influence the information that surfaces in search results. As a consequence, the chat algorithms may prioritize sources that align with the prevailing narrative, emphasizing the benefits and downplaying potential risks associated with hormonal contraceptives. This can inadvertently lead to an incomplete and skewed understanding of the topic, as critical perspectives and studies highlighting the risks may be overshadowed or marginalized.

These biases can be further compounded by political and allied interests that seek to shape the discourse surrounding birth control. Various stakeholders could attempt to manipulate search results, either directly or indirectly, to direct the users toward their agendas. As a result, individuals increasingly relying on chatbot conversations may struggle to access well-rounded and unbiased information about the potential health risks associated with oral contraceptive pills.

This imbalance in the available literature underscores the importance of critically evaluating information obtained through these chatbots. It highlights the need for individuals to be aware of the biases that can be inherent in search results and to actively seek out diverse sources of information. By consulting reputable scientific journals, academic research databases, and trusted healthcare resources, individuals can obtain a more comprehensive understanding of the risks and benefits associated with oral contraceptive pills.

Moreover, this example demonstrates the limitations of relying solely on internet searches for accessing nuanced information on public health topics. It emphasizes the significance of seeking guidance from healthcare professionals who possess the expertise to navigate and interpret scientific literature objectively. Engaging in open and informed discussions with healthcare providers allows individuals to receive personalized advice, address specific concerns, and obtain a more holistic view of the risks and benefits of oral contraceptive pills.

Discussion:

Chatbots operate based on sophisticated algorithms that analyze user queries and generate responses. However, these algorithms are not immune to biases, as they are likely to be designed to prioritize certain information sources and viewpoints. For example, in the case of oral contraceptives, the algorithm may favor sources that emphasize the benefits while downplaying or omitting information about potential risks associated with their use, especially when their use may be desired by public health agencies for the betterment of women and reproductive health. For example, looking at the increased risk of developing cancer over use for a long period of time especially in vulnerable demographics (certain ethnicity). This is based on a recent user experience. Each user experience may surely vary.  Consequently, the chatbot may provide incomplete or skewed information, hindering individuals’ ability to obtain a comprehensive understanding of the topic.

Moreover, the bias encountered in chatbot responses can hinder the retrieval of scientific papers and systematic reviews that delve into the potential risks associated with oral contraceptives. These studies may present nuanced findings, highlighting adverse effects, contraindications, or specific populations for whom caution is advised. However, due to the bias toward promoting contraceptive use, the chatbot may overlook or underrepresent such studies, limiting individuals’ exposure to critical information. Various factors such as optimization techniques, sponsored content, and commercial interests can influence the visibility and ranking of information, potentially skewing the presentation of viewpoints. In the context of public health, biases can significantly impact the availability and accessibility of information related to oral contraceptive pills and their associated health risks.

In navigating information gleaned from such chatbots, it is imperative for individuals to exercise critical thinking and evaluate search results meticulously. A casual conversation with a chatbot may not always yield a balanced view, as the algorithms are likely to prioritize certain sources and perspectives. Furthermore, the ”best interests” of the public, as determined by the algorithms (in turn determined by interests that be), may not align with providing comprehensive and unbiased information. It is essential to be aware of these limitations and actively seek out diverse sources of information.

Policy recommendations

Enhanced transparency and disclosure

• Advocate for makers of Chatbots like ChatGPT, Bard, etc. to provide more transparency regarding the factors influencing ‘fact presentation’ and visibility of information.
• Encourage Chatbot makers to disclose potential conflicts of interest, sponsorships, or biases that may impact search results.

Promoting Critical Health Literacy

• Advocate for the integration of critical evaluation and information literacy skills into chatbot interactions, empowering individuals to critically assess online health information.
• Collaborate in the development of educational campaigns and resources that educate the public about biases in chatbot responses and strategies for effectively navigating and evaluating the information provided by chatbots.

Collaboration between Public Health Experts and Tech Companies

• Foster partnerships between public health experts and technology companies to ensure the development of search algorithms that prioritize the presentation of balanced, evidence-based information.
• Engage in ongoing dialogue to address concerns related to search result biases and work towards optimizing the retrieval of reliable health information.

Conclusions:

Bias in public health-related conversations with chatbots such as ChatGPT or Bard or any emerging ones poses significant challenges to individuals seeking accurate and comprehensive information. By acknowledging the existence of biases and actively addressing them, we can foster a digital landscape that enables individuals to make informed decisions about their health. Through policy recommendations such as enhanced transparency, promoting critical health literacy, and collaboration between public health experts and tech companies, we can mitigate the impact of bias and ensure equitable access to reliable information. Empowered by critical evaluation skills, individuals can navigate public health internet searches with confidence, unveiling hidden truths and making informed choices that contribute to their overall well-being.

References:

Ghosal, Deepanway, Navonil Majumder, Soujanya Poria, Niyati Chhaya, and Alexander Gelbukh. 2019. “DialogueGCN: A Graph Convolutional Neural Network for Emotion Recognition in Conversation”. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). Association for Computational Linguistics. https://doi.org/10.18653/v1/d19-1015. 

Drs. Natarajan Ganesan and Thanemozhi G. Natarajan contributed to this article. 

During the preparation of this work, the authors used grammar-checking tools and generative AI in order to improve the readability and organization of the content. After using this tool/service, the authors reviewed and edited the content as needed, and take full responsibility for the content presented.

Building a Sequencing Data Analysis Platform: A Roadmap and Strategy for Success

Introduction:

With the rise of next-generation sequencing technologies, there is an increasing demand for efficient and user-friendly data analysis platforms. Researchers and organizations require powerful and flexible tools to analyze their sequencing data, extract insights, and make informed decisions. In this article, we will discuss a product roadmap and strategy for building a sequencing data analysis platform that can meet these needs.

Product Roadmap:

Phase 1: Initial Development

The first phase of building a sequencing data analysis platform is to develop the core functionality. This includes building a user-friendly interface, implementing basic data import and processing functionality, developing a pipeline for basic quality control and filtering of raw data, and integrating popular bioinformatics tools for read mapping and variant calling.

The user interface should be designed to be intuitive and easy to use, allowing users to navigate through the platform effortlessly. The platform should have basic data processing capabilities, such as handling raw data files and converting them into usable formats. Quality control and filtering of raw data should be implemented to ensure that the data is of sufficient quality for downstream analysis. Finally, the integration of popular bioinformatics toolsfor read mapping and variant calling is necessary to provide a comprehensive analysis of the sequencing data.

Phase 2: Feature Expansion

The second phase of building a sequencing data analysis platform involves expanding the platform's functionality. This includes adding data visualization and exploration options, implementing more advanced quality control and filtering options, developing additional pipelines for specific analysis types (e.g. RNA-seq, ChIP-seq), and integrating machine learning algorithms for predictive analysis.

Data visualization and exploration are essential for understanding complex data, making it easy for users to extract meaningful insights from their sequencing data. Advanced quality control and filtering options should be implemented to enable users to customize their data processing pipeline based on their research needs. The development of additional pipelines for specific analysis types, such as RNA-seq and ChIP-seq, will expand the platform's applicability to a broader range of research fields. Finally, the integration of machine learning algorithms can provide predictive analysis capabilities, enabling users to make more informed decisions based on their data.

Phase 3: Scaling and Integration

The third and final phase of building a sequencing data analysis platform involves scaling and integration. This includes optimizing the platform for scalability and cloud deployment, developing APIs for integration with other bioinformatics tools and workflows, offering customization options for advanced users, and providing support and training for users.

Optimizing the platform for scalability and cloud deployment is essential to ensure that the platform can handle large datasets and can be easily accessed from anywhere in the world. The development of APIs will enable the platform to integrate with other bioinformatics tools and workflows, providing a seamless experience for users. Offering customization options for advanced users, such as the ability to develop and integrate their own analysis pipelines, will enable them to tailor the platform to their specific needs. Finally, providing support and training for users is crucial to ensure that they can fully utilize the platform and achieve their research goals.

Strategy:

The strategy for building a successful sequencing data analysis platform involves identifying target users and their needs, building a user-friendly interface, implementing robust data processing, developing advanced analysis features, scaling and integrating the platform, and providing support and training for users.

Identifying target users and their needs is the first step in developing a successful sequencing data analysis platform. Understanding the types of researchers and organizations that would benefit from the platform and gathering feedback on their needs and pain points is crucial to ensure that the platform meets their requirements.

Building a user-friendly interface is essential to ensure that the platform is accessible and usable for all users. The platform should be designed with the user in mind, with intuitive navigation, clear labeling, and helpful tooltips. Implementing robust data processing capabilities, developing advanced analysis features, scaling and integrating the platform, and providing support and training for users are also key components of a successful sequencing data analysis platform. By following this roadmap and strategy, you can build a sequencing data analysis platform that meets the needs of researchers and organizations, enables them to extract valuable insights from their data, and accelerates scientific discovery.

  

mRNA vaccine vs other vaccines – Some FAQs

It is late September of 2020 and the Covid19 Wuhan Coronavirus has claimed nearly a million lives in a matter of months. US alone has ended with more than 200 thousand deaths so far, and there are little signs of the death toll abating in the near future. 

The race for vaccine was started right at earnest and progressed through different phases at an unprecedented pace, most possibly over pacing the countless rigors in the process. Given the scale of the pandemic, much of this is to be expected. Among the candidates to make the final cut are the mRNA vaccines, a new type of engineered vaccine hitherto untested extensively in humans. As a result, a lot of questions have come up regarding its safety and efficacy. Here are some of them …

1. What are some benefits and risks of an mRNA vaccine?

The benefits of an RNA vaccine (or mRNA vaccine, as it may be referred to as) stems mainly from the fact that the antigen coding transcripts are directly used by the host cellular machinery to be translated into active components, against which the antibodies shall be raised (by the body’s immune system) (1).

Figure 1: Two categories of mRNA constructs are being actively evaluated. Source: “The promise of mRNA vaccines: a biotech and industrial perspective” – npj Vaccines - https://www.nature.com/articles/s41541-020-0159-8/figures/1

Using this approach has a key advantage. The genetic material in the vaccine doesn’t have to enter the cell’s nucleus and incorporate it into the genome. Instead, it directly uses the translating machinery that converts the transcript into a protein. It is also possible to fine-tune the mRNA transcript by doing some chemical modifications to avoid degradation. 

To use an analogy, you use the ‘printer’ of the cell to print out pamphlets about the bad guy, rather than sending the ‘entire file’ to the cell’s ‘computer’.

The risks are yet to be known in humans since this approach fairly new in clinical practice. It is possible for the mRNA vaccine to induce some unintended immune responses including mRNA-Cargo interactions during the formulation process (2). There is no animal model as yet that matches perfectly with the human responses. 

Another possible concern could be that some m-RNA based vaccine platforms induce some potent type-I interferon responses. These have been known to be associated with inflammation and autoimmune responses. Therefore identifying individuals at increased risk for this would be key to administering this type of vaccine (3).

2. What about any side effects that are concerning? How common are these in such vaccine trials?

In the latest development, researchers at the Massachusetts General Hospital have identified some markers that may predict coagulation assisted complications in patients with COVID-19 (4). The risks for an mRNA based vaccine come with its own flavors besides other usual ones typically associated with vaccine trials. How significantly would they pan out as compared to other types of vaccine systems, remains yet to be seen.

According to the Moderna company’s website (5), their mRNA-1273 vaccine is against the stabilized form of Spike protein (S). Though this portion has remained fairly stable small variants have observed (6). Also, the vaccine is under the assumption that the molecular pathogenesis is from this protein alone. The virus is fast evolving. 

Also, the progression through different phases has been greatly accelerated so far, given the nature of the pandemic. As a result, many approval steps may have had to be rushed through, for expediency; one may have to assume that all these steps must have been duly vetted out as well.

3. What makes up for a very promising vaccine trial and convince someone to sign up for this trial vs. other vaccine trials?

The website of the company should have a clear and transparent timeline of the progress of the vaccine development so far; starting from their initial procurement of the sequence data and the contract date till today. It would be therefore possible to get an idea of the way in which the results are going to pan out, assuming the site continues to keep it that way (7). 

4. What is the ideal sample size for a Phase 3 study of such a vaccine trial? Can numbers alone suffice?

A couple of recent ongoing trials (8,9) have decided to enroll about 30,000 for their Phase 3 of clinical trials, and this is a fairly good enough for a phase trial of this size. Most of such trials run into few thousand anyway. The greater question should be if there is time enough for follow up studies to study the efficacy and safety (10) and observe for any adverse events. Given the current nature of the pandemic that again may have to be ‘accelerated’. How that may pan out to assuage the concerns of the public before public release is a different matter.

5. What does this Phase 3 study need to prove that it is a good vaccine?

Firstly, the vaccine has to be shown to be widely effective, matching up with beyond the success rates of the other vaccine trials. In addition to this, adverse events must be very insignificant in comparison to the rates for other vaccine therapies. All of these results must be publicly available and transparent, of course.

6. What is ‘vaccine efficacy’ and how would a ‘50% effective’ vaccine match up to one that is 70% effective? Can we expect one that is 70% or more effective?

An informal search on the web shows any trial success above 50% is good enough, 60% would really pushing it. I am not aware of 70% success rates, great if it does. A recent study by researchers from MIT looked into the success rates of Vaccine trials vs therapeutic trials and found that they were more likely to hit the dust than their non-vaccine counterparts BUT that reason has been mainly attributed to lack of investment and efforts, in the recent decades, in finding the ideal approaches to delivering vaccines effectively (11,12).

Numbers aside, at this point the aim should be mainly to increase the absolute numbers who are successfully immunized and can thus add up to the herd immunity. 

Also, this being an mRNA vaccine, the first of its kind in a worldwide major trial, is likely to be recorded in itself. Given the theory behind the vaccine, the success rate can be expected to be pretty high like its other counterparts.

7. What about some common fears, misconceptions, or misplaced logic?

Despite the year being 2020, the fears and myths against vaccines have grown so exponentially in the last few years that one would wonder … Well! There are any number of myths and all of them debunked as well. 

One of such misplaced logic, of particular relevance to the current pandemic, is "allowing herd immunity to do its job"! This is another way to skip vaccines under some pretext at a great cost to public health. To put it simply, it sounds like a game of Life-roulette.  Consider this! We have witnessed 200 thousand deaths in a matter of months and still counting!!! The US has probably never seen anything like this. As if gambling with one’s life were not enough you are just endangering everyone out there.  There are countless instances of people resorting to such an approach and regretting it badly. Not worth it! On the other hand, there are nations that have successfully taken active measures in this regard. One should learn from them and allow Science to do its job.

8. Why is it important to volunteer? How to allay fears of the risks?

The success of a clinical trial exclusively depends on the active participation of healthy volunteers. Consider this! If you can get a thrill out a rollercoaster OR bungee jump then this is far better; you become a hero helping science advance further for public health. In an open and transparent setting of a clinical trial, the volunteer will and should have access to all that entails the participation.