Patient Segmentation Using Data Mining Techniques

Srinivasan Sundararajan

Srinivasan Sundararajan

Patient Segmentation & Quality Patient Care

As the need for quality and cost-effective patient care increases, healthcare providers are increasingly focusing on data-driven diagnostics while continuing to utilize their hard-earned human intelligence. Simply put, data-driven healthcare is augmenting the human intelligence based on experience and knowledge.

Segmentation is the standard technique used in Retail, Banking, Manufacturing, and other industries that needs to understand their customers to provide better customer service. Customer segmentation defines the behavioral and descriptive profiles of customers. These profiles are then used to provide personalized marketing programs and strategies for each group.

In a way, patients are like customers to healthcare providers. Though the element of quality of care takes precedence than profit-making intention, a similar segmentation of patients will immensely benefit the healthcare providers, mainly for the following reasons:

  • Customizing the patient care based on their behavior profiles
  • Enabling a stronger patient engagement
  • Providing the backbone for data-driven decisions on patient profile
  • Performing advanced medical research like launching a new vaccine or trial

The benefits are obvious and individual hospitals may add more points to the above list; the rest of the article is about how to perform the patient segmentation using data mining techniques.

Data Mining for Patient Segmentation

In Data Mining a, segmentation or clustering algorithm will iterate over cases in a dataset to group them into clusters that contain similar characteristics. These groupings are useful for exploring data, identifying anomalies in the data, and creating predictions. Clustering is an unsupervised data mining (machine learning) technique used for grouping the data elements without advance knowledge of the group definitions.

K-means clustering is a well-known method of assigning cluster membership by minimizing the differences among items in a cluster while maximizing the distance between clusters. Clustering algorithm first identifies relationships in a dataset and generates a series of clusters based on those relationships. A scatter plot is a useful way to visually represent how the algorithm groups data, as shown in the following diagram. The scatter plot represents all the cases in the dataset, and each case is a point on the graph. The cluster points on the graph illustrate the relationships that the algorithm identifies.

AIOps Artificial Intelligence for IT Operations

One of the important parameters for a K-Means algorithm is the number of clusters or the cluster count. We need to set this to a value that is meaningful to the business problem that needs to be solved. However, there is good support in the algorithm to find the optimal number of clusters for a given data set, as explained next.

To determine the number of clusters for the algorithm to use, we can use a plot of the within cluster’s sum of squares, by the number of clusters extracted. The appropriate number of clusters to use is at the bend or ‘elbow’ of the plot. The Elbow Method is one of the most popular methods to determine this optimal value of k i.e. the number of clusters. The following code creates a curve.

AIOps Digital Transformation Solutions
AI Devops Automation Service Tools

In this example, based on the graph, it looks like k = 4 would be a good value to try.

Reference Patient Segmentation Using K-Means Algorithm in GAVS Rhodium Platform

In GAVS Rhodium Platform, which helps healthcare providers with Patient Data Management and Patient Data Sharing, there is a reference implementation of Patient Segmentation using K-Means algorithm. The following are the attributes that are used based on a publicly available Patient admit data (no personal information used in this data set). Again in the reference implementation sample attributes are used and in a real scenario consulting with healthcare practitioners will help to identify the correct attributes that is used for clustering.

 To prepare the data for clustering patients, patients must be separated along the following dimensions:

  • HbA1c: Measuring the glycated form of hemoglobin to obtain the three-month average of blood sugar.
  • Triglycerides: Triglycerides are the main constituents of natural fats and oils. This test indicates the amount of fat or lipid found in the blood.
  • FBG: Fasting Plasma Glucose test measures the amount of glucose levels present in the blood.
  • Systolic: Blood Pressure is the pressure of circulating blood against the walls of Blood Vessels. This test relates to the phase of the heartbeat when the heart muscle contracts and pumps blood from the chambers into the arteries.
  • Diastolic: The diastolic reading is the pressure in the arteries when the heart rests between beats.
  • Insulin: Insulin is a hormone that helps move blood sugar, known as glucose, from your bloodstream into your cells. This test measures the amount of insulin in your blood.
  • HDL-C: Cholesterol is a fat-like substance that the body uses as a building block to produce hormones. HDL-C or good cholesterol consists primarily of protein with a small amount of cholesterol. It is considered to be beneficial because it removes excess cholesterol from tissues and carries it to the liver for disposal. The test for HDL cholesterol measures the amount of HDL-C in blood.
  • LDL-C: LDL-C or bad cholesterol present in the blood as low-density lipoprotein, a relatively high proportion of which is associated with a higher risk of coronary heart disease. This test measures the LDL-C present in the blood.
  • Weight: This test indicates the heaviness of the patient.

The above tests are taken for the patients during the admission process.

The following is the code snippet behind the scenes which create the patient clustering.

Best AIOps Platforms Software

The below is the output cluster created from the above algorithm.

Just from this sample, healthcare providers can infer the patient behavior and patterns based on their creatinine and glucose levels, in real-life situations other different attributes can be used.

AI will play a major role in future healthcare data management and decision making and data mining algorithms like K-Means provide an option to segment the patients based on the attributes which will improve the quality of patient care.

About the Author –

Srini is the Technology Advisor for GAVS. He is currently focused on Healthcare Data Management Solutions for the post-pandemic Healthcare era, using the combination of Multi Modal databases, Blockchain and Data Mining. The solutions aim at Patient data sharing within Hospitals as well as across Hospitals (Healthcare Interoprability), while bringing more trust and transparency into the healthcare process using patient consent management, credentialing and zero knowledge proofs.

Getting The Best From Healthcare AI

Tim perry

Tim Perry

Co-founder & CIO, Healthcare Too

Advisor to the CIO of AgFirst

Is Healthcare Artificial Intelligence The Answer?

To help explain the future of healthcare Artificial Intelligence (AI) let’s borrow a few lines from Lewis Carroll’s classic Alice in Wonderland:

Alice: Would you tell me, please, which way I ought to go from here?

The Cheshire Cat: That depends a good deal on where you want to get to.

So it is with healthcare AI. It really just depends on where we want to go with healthcare in the US (and globally for that matter). Much of the current conversation seems to be on using AI to improve medical care. Hospitals want to use data from retail clinics, homes, government agencies, and more to predict individual medical needs. Big Tech companies try to apply AI to diagnose diseases better than physicians. Insurers collect massive amounts of data to manage better their risk pool through AI.

AI in Healthcare

A common theme for so many of these healthcare AI scenarios is that AI improves the efficiency of the current system. That improvement is supposedly good for everyone: patients, providers, insurers. And that is also where we get it terribly wrong. If we really want to make the most of healthcare AI investments and promote wellbeing there are two things we must remember:

  1. No one wants to be a patient, but everyone wants to be healthy.
  2. AI offers only point solutions, not a universal truth.

Everyone Wants To Be Healthy

No one wants to be a patient, not even doctors and nurses. The patient experience is painful, frightening, and terribly expensive (in the US anyway). Everyone would much prefer to remain healthy and never see the inside of a hospital. In the US sick care system, however, there is a financial incentive only when there is a diagnosis and treatment. Healthcare AI solutions that do not produce more diagnoses and treatments are not viable in our current sick care system. Like Alice, we must know which way we want to go: more sick care or a new system for health and wellbeing?

AI Offers Only Point Solutions

Artificial Intelligence comes in two basic flavors: 1) General and 2) Narrow. Again, we must plan and invest knowingly to get to where we want to go. These investments over the next 5-10 years will largely determine the direction of Healthcare for decades.

General AI

This is the sexy AI, the stuff we see in science fiction. Computers are so smart that they can address any type of problem decisively and with lightning speed. We use words like “reasoning” or “thinking” when we imagine the power of General AI. As far as our investments and resources go for healthcare AI the General AI option is many years away. We cannot afford to invest in fiction.

Narrow AI

That leaves us to consider narrow AI. These are solutions that are focused on a specific task like search, image analysis, or driving a car. Each is a significant undertaking and requires advanced capabilities. These point solutions in healthcare AI are already underway. Unfortunately, many of the solutions are those that focus on more diagnoses and treatments in the current sick care model. This is not where we want to go.

Healthcare AI For Health

IT Operations Management Software

Focused on Narrow AI, we can envision healthcare where AI promotes health as a state of complete physical, mental, and social well-being and not merely the absence of disease or infirmity (as the World Health Organization defines health). There are near countless examples of improving health with AI when we think holistically about real healthcare requirements:

  • Instead of more diagnoses and treatments, what about healthcare AI that weans patients off medications with improvements in nutrition and other social determinants of health?
  • Maybe AI that offers an appropriate personalized spiritual thought based on facial expression, voice tone, or body posture?
  • What about AI for positive online social interactions that help filter negative experiences and protect privacy instead of tracking every movement/action to provide more ads?
  • If we allow AI-driven cars on our roads why not self-driving food trucks with fresh produce and prepared foods for areas we currently call “food deserts”?
  • And just imagine, if you will, an AI that evaluated a person’s current health not only against mountains of conventional medical data from the last hundred years but millennia of data from traditional medical systems like Ayurveda and Traditional Chinese Medicine?

There are countless applications for real healthcare AI. We only need to decide where we are going. Be Well!

About the Author –

Tim Perry, MPA, MS, CPHIMS, CISSP is the Co-Founder & Chief Information Officer of Consumer Health platform HealthCare Too. At present, Tim is an advisor to the CIO of AgFirst and plays a key role in Strategy and Planning of the organization. Over the past 3 decades, Tim has worked in Fortune 50 executive leadership roles as well as startups and has developed a deep passion for transforming healthcare. He is blessed with a wonderful wife and two inspiring children. Tim has practiced Tai Chi (Taiji Chuan) for 20 years and enjoys cooking wholesome (and easy) meals.

Palo Alto Firewall – DNS Sinkhole

Ganesh Kumar J

Starting with PAN-OS 6.0, DNS sinkhole is an action that can be enabled in Anti-Spyware profiles. A DNS sinkhole can be used to identify infected hosts on a protected network using DNS traffic in environments where the firewall can see the DNS query to a malicious URL.

The DNS sinkhole enables the Palo Alto Networks device to forge a response to a DNS query for a known malicious domain/URL and causes the malicious domain name to resolve to a definable IP address (fake IP) that is given to the client. If the client attempts to access the fake IP address and there is a security rule in place that blocks traffic to this IP, the information is recorded in the logs.

Sample Flow

We need to keep the following in mind before assigning an IP address to DNS sinkhole configuration.

When choosing a “fake IP”, make sure that the IP address is a fictitious IP address that does not exist anywhere inside the network. DNS and HTTP traffic must pass through the Palo Alto Networks firewall for the malicious URL to be detected and for the access to the fake IP to be stopped. If the fake IP is routed to a different location, and not through the firewall, this will not work properly.

Steps:

  1. Make sure the latest Antivirus updates are installed on the Palo Alto Networks device. From the WebUI, go to Device > Dynamic Updates on the left. Click “Check Now” in the lower left, and make sure that the Anti-Virus updates are current. If they are not, please do that before proceeding. The Automatic Updates can be configured if they are not setup.

Fig1.1

IT Automation with AI

Note: A paid Threat Prevention subscription for the DNS sinkhole is required to function properly.

  1. Configure the DNS Sinkhole Protection inside an Anti-Spyware profile. Click on the Objects > Anti-Spyware under Security Profiles on the left.
    Use either an existing profile or create a new profile. In the example below the “alert-all” is being used:

Fig1.2:

Office 365 Migration

Click the name of the profile – alert-all, click on the DNS Signatures tab.

Fig1.3:

Software Test Automation Platform

Change the “Action on DNS queries” to ‘sinkhole’ if it is not already set to sinkhole.
Click on the Sinkhole IPv4 field, either select the default Palo Alto Networks Sinkhole IP (72.5.65.111) or a different IP of your choosing. If you opt to use your own IP, ensure the IP is not used inside your network and preferably not routable over the internet (RFC1918).
Click on Sinkhole IPv6 and enter a fake IPv6 IP. Even if IPv6 is not used, something still needs to be entered. The example shows ::1. Click OK. 

Note: If nothing is entered for the Sinkhole IPv6 field, OK will remain grayed out.

  1. Apply the Anti-Spyware profile on the security policy that allows DNS traffic from the internal network (or internal DNS server) to the internet. Click on Policies> Security on the left side. Inside the rules, locate the rule that allows DNS traffic outbound, click on the name, go to the Actions tab, and make sure that the proper Anti-Spyware profile is selected. Click OK..

Fig1.4:

Software Product Engineering Services

  1. The last thing needed is to have a security rule that will block all web-browsing and SSL access to the fake IP 72.5.65.111 and also :1 if using IPv6. This will ensure to deny traffic to the fake IP from any infected machines.

Fig1.5:

Security Iam Management Tools

  1. Commit the configuration

Fig1.6:

Rpa in Infrastructure Management

(To be continued…)

References:

About the Author –

Ganesh is currently managing Network, Security and engineering team for a large US based customer. He has been associated with the Network & Security domain for more than 15 years.

Container Security

Anandharaj V

We live in a world of innovation and are beneficiaries of new advancements. New advancements in software technology also comes with potential security vulnerabilities.

‘Containers’ are no exception. Let us first understand what a container is and then the vulnerabilities associated with it and how to mitigate them.

What is a Container?

You might have seen containers in the shipyard. It is used to isolate different cargos which is transported via ships. In the same way, software technologies use a containerization approach.

Containers are different from Virtual Machines (VM) where VMs need a guest operating system which runs on a host operating system (OS). Containers uses OS virtualization, in which required processes, CPU, Memory, and disk are virtualized so that containers can run without a separate operating system.

In containers, software and its dependencies are packaged so that it can run anywhere whether on-premises desktop or in the cloud.

IT Infrastructure Managed Services

Source: https://cloud.google.com/containers

As stated by Google, “From Gmail to YouTube to Search, everything at Google runs in containers”.

Container Vulnerabilities and Countermeasures

Containers Image Vulnerabilities

While creating a container, an image may be patched without any known vulnerabilities. But a vulnerability might have been discovered later, while the container image is no longer patched. For traditional systems, it can be patched when there is a fix for the vulnerability without making any changes but for containers, updates should be upstreamed in the images, and then redeployed. So, containers have vulnerabilities because of the older image version which is deployed.

Also, if the container image is misconfigured or unwanted services are running, it will lead to vulnerabilities.

Countermeasures

If you use traditional vulnerability assessment tools to assess containers, it will lead to false positives. You need to consider a tool that has been designed to assess containers so that you can get actionable and reliable results.

To avoid container image misconfiguration, you need to validate the image configuration before deploying.

Embedded Malware and Clear Text Secrets

Container images are collections of files packaged together. Hence, there are chances of malicious files getting added unintentionally or intentionally. That malicious software will have the same effect as of the traditional systems.

If secrets are embedded in clear text, it may lead to security risks if someone unauthorized gets access.

Countermeasures

Continuous monitoring of all images for embedded malware with signature and behavioral detection can mitigate embedded malware risks.

 Secrets should never be stored inside of containers image and when required, it should be provided dynamically at runtime.

Use of Untrusted Images

Containers have the advantages of ease of use and portability. This capability may lead teams to run container images from a third party without validating it and thus can introducing data leakage, malware, or components with known vulnerabilities.

Countermeasures

Your team should maintain and use only trusted images, to avoid the risk of untrusted or malicious components being deployed.

Registry Risks

Registry is nothing but a repository for storing container images.

  1. Insecure connections to registries

Images can have sensitive information. If connections to registries are performed over insecure channels, it can lead to man-in-the-middle attacks that could intercept network traffic to steal programmer or admin credentials to provide outdated or fraudulent images.

You should configure development tools and containers while running, to connect only over the encrypted medium to overcome the unsecured connection issue.

  1. Insufficient authentication and authorization restrictions

As we have already seen that registries store container images with sensitive information. Insufficient authentication and authorization will result in exposure of technical details of an app and loss of intellectual property. It also can lead to compromise of containers.

Access to registries should authenticated and only trusted entities should be able to add images and all write access should be periodically audited and read access should be logged. Proper authorization controls should be enabled to avoid the authentication and authorization related risks.

Orchestrator Risks

  1. Unbounded administrative access

There are many orchestrators designed with an assumption that all the users are administrators but, a single orchestrator may run different apps with different access levels. If you treat all users as administrators, it will affect the operation of containers managed by the orchestrator.

Orchestrators should be given the required access with proper role-based authorization to avoid the risk of unbounded administrative access.

  1. Poorly separated inter-container network traffic

In containers, traffic between the host is routed through virtual overlay networks. This is managed by the orchestrator. This traffic will not be visible to existing network security and management tools since network filters only see the encrypted packets traveling between the hosts and will lead to security blindness. It will be ineffective in monitoring the traffic.

To overcome this risk, orchestrators need to configure separate network traffic as per the sensitivity levels in the virtual networks.

  1. Orchestrator node trust

You need to give special attention while maintaining the trust between the hosts, especially the orchestrator node. Weakness in orchestrator configuration will lead to increased risk. For example, communication can be unencrypted and unauthenticated between the orchestrator, DevOps personnel, and administrators.

To mitigate this, orchestration should be configured securely for nodes and apps. If any node is compromised, it should be isolated and removed without disturbing other nodes.

Container Risks

  1. App vulnerabilities

It is always good to have a defense. Even after going through the recommendations, we have seen above; containers may still be compromised if the apps are vulnerable.

As we have already seen that traditional security tools may not be effective when you use it for containers. So, you need a container aware tool which will detect behavior and anomalies in the app at run time to find and mitigate it.

  1. Rogue containers

It is possible to have rogue containers. Developers may have launched them to test their code and left it there. It may lead to exploits as those containers might not have been thoroughly checked for security loopholes.

You can overcome this by a separate environment for development, test, production, and with a role-based access control.

Host OS Risks

  1. Large attack surface

Every operating system has its attack surface and the larger the attack surface, the easier it will be for the attacker to find it and exploit the vulnerability and compromise the host operating system and the container which run on it.

You can follow the NIST SP 800-123 guide to server security if you cannot use container specific operating system to minimize the attack surface.

  1. Shared kernel

If you only run containers on a host OS you will have a smaller attack surface than the normal host machine where you will need libraries and packages when you run a web server or a database and other software.

You should not mix containers and non-containers workload on the same host machine.

If you wish to further explore this topic, I suggest you read NIST.SP.800-190.


References

About the Author –

Anandharaj is a lead DevSecOps at GAVS and has over 13 years of experience in Cybersecurity across different verticals which include Network Security, application Security, computer forensics and cloud security.

Customer Focus Realignment in a Pandemic Economy

Ashish Joseph

Business Environment Overview

The Pandemic Economy has created an environment that has tested businesses to either adapt or perish. The atmosphere has become a quest for the survival of the fittest. On the brighter side, organizations have stepped up and adapted to the crisis in a way that they have worked faster and better than ever before. 

During this crisis, companies have been strategic in understanding their focus areas and where to concentrate on the most. From a high-level perspective, we can see that businesses have focused on recovering the sources of their revenues, rebuilding operations, restructuring the organization, and accelerating their digital transformation initiatives. In a way, the pandemic has forced companies to optimize their strategies and harness their core competencies in a hyper-competitive and survival environment.

Need for Customer Focused Strategies

A pivotal and integral strategy to maintain and sustain growth is for businesses to avoid the churn of their existing customers and ensure the quality of delivery can build their trust for future collaborations and referrals. Many organizations, including GAVS, have understood that Customer Experience and Customer Success is consequential for customer retention and brand affinity. 

Businesses should realign themselves in the way they look at sales funnels. A large portion of the annual budget is usually allocated towards the top of the funnel activities to acquire more customers. But companies with customer success engraved in their souls, believe in the ideology that the bottom of the funnel feeds the top of the funnel. This strategy results in a self-sustaining and recurring revenue model for the business.

An independent survey conducted by the Customer Service Managers and Professionals Journal has found that companies pay 6x times more to acquire new customers than to keep an existing one. In this pandemic economy, the costs for customer acquisition will be much higher than before as organizations must be very frivolous in their spending. The best step forward is to make sure the companies strive for excellence in their customer experience and deliver measurable value to them. A study conducted by Bain and Company titled “Prescription for Cutting Costs” talks about how increasing customer retention by 5% increases profits from 25%-95%. 

The path to a sustainable and high growth business is to adopt customer-centric strategies that yield more value and growth for its customers. Enhancing customer experience should be prime and proper governance must be in place to monitor and gauge strategies. Governance in the world of the customer experience must revolve around identifying and managing resources needed to drive sustained actions, establishing robust procedures to organize processes, and ensuring a framework for stellar delivery.

Scaling to ever-changing customer needs

A research body called Walker Information conducted an independent research on B2B companies focusing on key initiatives that drive customer experiences and future growth. The study included various customer experience leaders, senior executives, and influencers representing a diverse set of business models in the industry. They published the report titled “Customer 2020: A Progress Report” and the following are strategies that best meet the changing needs of customers in the B2B landscape.

AI Devops Automation Service Tools

Over 45% of the leaders highlighted the importance of developing a customer-centric culture that simplifies products and processes for the business. Now the question that we need to ask ourselves is, how do we as an organization scale up to these demands of the market? I strongly believe that each of us, in the different roles we play in the organization, has an impact.

The Executive Team can support more customer experience strategies, formulate success metrics, measure the impact of customer success initiatives, and ensure alignment with respect to the corporate strategy.

The Client Partners can ensure that they represent the voice of the customer, plot a feasible customer experience roadmap, be on point with customer intelligence data, and ensure transparency and communication with the teams and the customers. 

The cross-functional team managers and members can own and execute process improvements, personalize and customize customer journeys, and monitor key delivery metrics.

When all these members work in unison, the target goal of delivery excellence coupled with customer success is always achievable.

Going Above and Beyond

Organizations should aim for customers who can be retained for life. The retention depends upon how much a business is willing to go the extra mile to add measurable value to its customers. Business contracts should evolve into partnerships that collaborate on their competitive advantages that bring solutions to real-world business problems. 

As customer success champions, we should reevaluate the possibilities in which we can make a difference for our customers. By focusing on our core competencies and using the latest tools in the market, we can look for avenues that can bring effort savings, productivity enhancements, process improvements, workflow optimizations, and business transformations that change the way our customers do business. 

After all, We are GAVS. We aim to galvanize a sense of measurable success through our committed teams and innovative solutions. We should always stride towards delivery excellence and strive for customer success in everything we do.

About the Author –

Ashish Joseph is a Lead Consultant at GAVS working for a healthcare client in the Product Management space. His areas of expertise lie in branding and outbound product management.

He runs a series called #BizPective on LinkedIn and Instagram focusing on contemporary business trends from a different perspective. Outside work, he is very passionate about basketball, music, and food.