Corrective action/preventive action (CAPA) is the proverbial thorn in the side of quality professionals. Every day they deal with whether to open a CAPA, investigating it once opened, creating and tracking action items, ensuring the CAPA is on track, and hunting down bottlenecks in the process. It can be a months-long process to find the root cause for the event you’re investigating and to enact effective preventive and corrective actions. If you can find the root cause at all.
Most problems don’t solve themselves, but this one has the potential to do just that. Using advanced data analytics and artificial intelligence (AI) applications, pharmaceutical companies can automate much of the CAPA process. Just digitizing so that all the information is in one place is a big step forward. But once you implement programs that can analyze that data, you save even more time. And by using machine learning (ML), a subset of AI, the program gets smarter, faster, and more accurate the longer it operates. Eventually, the program can detect, investigate and resolve a CAPA with little or no human involvement.
Laying the foundation
While the idea of a self-investigating CAPA is certainly possible, there are a lot of steps a company needs to take before they can get there. As mentioned, ML or any type of AI needs to be trained using data. And if that data is incomplete or incorrect, the application won’t perform properly.
When it comes to CAPA, pertinent data comes from systems throughout the product lifecycle. Supplier, manufacturing, quality event, and postmarket data can all play a part in CAPA. The program needs access to data about a defective part or nonconforming materials from a supplier, problems on the manufacturing line, previously closed CAPAs, and customer complaints to come to the proper conclusions. The root cause of the event you are investigating could be related to any of these issues, but that can be difficult to determine when conducting a typical manual CAPA investigation. AI and analytics systems can analyze that data, find the relationships and causal factors, and point you toward the root cause.
Fortunately, this doesn’t mean scrapping every system and process already in place and investing in a single, all-powerful solution. Actually, companies can continue to use the systems they have in place as long as they can integrate them. Integrating operations between enterprise resource planning (ERP), manufacturing execution systems (MES), customer relationship management (CRM), and other solutions the company uses lets the systems share data and prevents data entry errors caused by manually reentering information.
This interconnectivity ensures that analytics or AI have complete data to work with, leading to accurate conclusions. With the right program, quality professionals can track down the root cause and avoid future problems with predictive analytics.
Automated quality event management
“Quality event” is usually a euphemism for “something that went wrong.” In some cases, “something that went horribly wrong.” Problems and issues are inevitable, but that doesn’t mean they’re unpredictable or uncontrollable. It also doesn’t mean that every quality event warrants a CAPA. The question of whether to open a CAPA is among the biggest pain points for quality professionals. Unfortunately, in some organizations, the default is to open a CAPA for fear of missing something. Yet, analytics tools can help enormously in making this decision. In fact, they can make the decision for you.
One example of this comes from monitoring customer complaints. Some of these will be minor, but some are indicative of serious adverse events. If an analytics solution is set up correctly, it can help pharma companies avoid those adverse events. Natural language processing (NLP), another member of the AI family, can monitor complaints about trends and anomalies. A company can set a threshold so that when a certain number of similar complaints about a particular product are received, the system automatically opens a CAPA. In some cases, the investigation will still require human involvement. In other cases, the entire CAPA process can be left in the hands of an AI.
Many CAPA investigations are about what has already gone wrong. Quality professionals spend so much time putting out fires that preventive action can be neglected. This is an area where predictive analytics can excel. Think of how much time and trouble you could save if you had indications beforehand that an adverse event might occur. Once the system understands what factors in the data have been associated with prior adverse events (e.g., batches with test results in a certain range, deviating from processes in particular ways, a particular material, complaints of specific issues, etc.), the system can then begin to identify high-risk batches for you and point out where additional testing and quality checking might be helpful.
Data analytics and AI offer benefits to QEM and CAPA, but those processes interact with other processes in the organization. The same technology offers improvements in those areas as well.
Analytics-enabled risk management
So much of risk management is understanding how one thing will affect another. And humans aren’t always very good at that. We confuse correlation and causation and our own biases skew how we interpret data. Using predictive analytics can alleviate or eliminate some of these problems. Every corrective or preventive action has an effect, which can entail risk. Recognizing those risks and properly mitigating them is vital. A common way of dealing with risk is a failure mode and effects analysis (FMEA). Unfortunately, for many companies, risk management ends once the FMEA is created.
An analytics solution can continually incorporate new inputs and determine the results. For example, the program can tell you if a process change will introduce a new hazard and the risks that the hazard poses. It can further tell you how to mitigate that problem. Analytics-enabled risk management connects the dots between actions and reactions in a way that manual methods can’t. Predictive analytics can identify trends and anomalies before they become large problems.
A good example of this recalls. Based on customer complaints and adverse events, a manufacturer may choose to issue a recall. A recall in this light is expensive, time-consuming, and creates a bad image for the company. When ML, NLP, and analytics are employed, the need for a recall can be detected sooner. The earlier the problem is detected, the sooner the recall can begin, which lessens the expense for the company. Even better, early detection can prevent the defective product from being sent out in the first place.
Predicting manufacturing needs
Process improvement and optimization frequently come back to the manufacturing line. Corrective action may involve a change to a work instruction or equipment calibration. And preventive action can involve alerts to detect when an employee needs training or indicate a need for increased testing.
Some of these efficiencies are more basic than analytics or AI. For example, a manufacturing system that’s connected to document management could automatically update the work instruction (WI) when a standard operating procedure (SOP) changes. The potential of analytics is much larger than this, to the point that it can anticipate when a change to an SOP might be needed, what that change should be, and what effect it will have. For example, if you commonly see process deviations from operators who were all trained using a particular SOP or WI, analytics could suggest taking a closer look at that training to see what might be clarified or improved.
Predictive analytics tools are useful for ensuring you meet your production needs. Programs can anticipate a shortage far in advance of when a human would notice it. Similarly, they can determine how to respond to a dramatic increase or decrease in demand. Using information from batch records, an analytics solution can tell you which employees should be included on a team to increase yield by a certain percent. It can also determine which materials from a particular supplier tend to result in fewer defects or higher yield.
Take the first step
This article largely focused on CAPA and quality events, but analytics and data-based decisions can improve any area of a business. The data you need for these decisions should be coming from all over the organization, so it makes sense that a solution would also be useful companywide. It’s easy to see the benefits for quality and manufacturing, but there are definitely use cases outside of those departments. An important first step is showing employees how the technology will make their lives easier — and not replace them. Of course, the first employees you should convince are in the C-suites.
It’s difficult to do anything in a company without executive buy-in. It helps to connect any change to the bottom line. Find out what key performance indicators (KPIs) are most important to your company. There really aren’t many KPIs that analytics and/or AI initiatives wouldn’t help. These are some dots that you’re perfectly capable of connecting without software, but you’ll still need data. Start with what your company is capable of right now. Then look at analytics solutions that would improve those KPIs. From there you can calculate savings in cost and time, and improvements in right-first-time metrics.
Nearly every day a new article describing the vast potential of AI is published. But it’s hard to know how those big claims would help you in the work you do every day. AI, and its less glamorous cousin data analytics, are a goldmine when it comes to improving the CAPA process. Quality professionals don’t typically enjoy running around putting out fires all day, but they don’t always have time for anything else. With the right level of analytics, you’ll be dealing with fewer CAPAs because they’ll automatically be resolved or investigated with minimal involvement on your part. You’ll also have fewer issues that escalate to the point that they need a CAPA.