Enhancing Data Security Resilience in AI-Driven Digital Transformation: Exploring Industry Challenges and Solutions Through ALCOA+ Principles

Abstract

Background:

The Medicines and Healthcare Products Regulatory Agency (MHRA) defines data integrity as the maintenance of accuracy, consistency, and completeness of data over time. Recently, “artificial intelligence” has become prevalent across industries, education, culture,and technology, denoting systems that mimic human intelligence and critical thinking using computers and related technologies.

Objective:

This article examines the construction of a robust artificial intelligence (AI) system and the incorporation of ALCOA+ principles for data validation, with a specific focus on enhancing data certainty and security.

Methods:

This study was carried out through a comprehensive review of various Scopus-indexed literature over the past decade. Results and Discussion: AI has been widely applied in Manufacturing System Optimization, involving organizing production systems, including machines, robots, conveyors, and related operations like maintenance and material handling. Moreover, it’s used for Process Monitoring, Diagnostics, and Prognostics in medicine, as well as supervision and regulation in industries. Yet, it’s not immune to shortcomings, which could result in system biases and jeopardize data security.

Conclusion:

This article explores the creation of a robust AI system, implementing ALCOA+ for data validation in AI-Driven Digital Transformation to improve data certainty and security in industries. It involves systematically recording AI system activities, ensuring database validity, sustaining data recording practices, regularly updating records, ensuring authenticity and completeness, and facilitating data accessibility for review and audits. As AI integration in education advances, there’s a crucial need for oversight to maintain data integrity in these systems.

1. BACKGROUND

The primary rationale for the U.S. Food and Drug Administration’s 21 CFR Part 11 is data integrity, which is essential to regulatory compliance (FDA). Since the FDA’s first guideline was published in 1963, the European Union (EU) and the FDA have released many recommendations on a range of subjects pertaining to data integrity for the pharmaceutical business. Regulators sought to ensure that the pharmaceutical sector collected reliable data throughout the drug development lifecycle and throughout commercialization. This is because a growing number of inspectors worldwide have recently issued warning letters on data integrity (1).

Data integrity, as defined by the Medicines and Healthcare Products Regulatory Agency (MHRA), is the degree to which all data are accurate, consistent, and complete during the course of their existence. On the other hand, the U.S. FDA defines data integrity as the correctness, consistency, and completeness of data. Complete, consistent, and accurate data should be traceable, legible, contemporaneously documented, original or a genuine copy, and accurate (ALCOA). Furthermore, the concepts of Complete, Consistent, Enduring, and Available have been added to ALCOA, expanding it into a new chapter called ALCOA+ (2).

Data integrity is consistently implemented in all stages of its implementation. Clinical trials use several methodologies, including Risk Based Monitoring (RBM), which ensures the protection of human participants, enhances the accuracy of data, and minimises expenses associated with drug development. Furthermore, the implementation of data monitoring via corrective actions may promptly identify faults in their initial stages, therefore averting the occurrence of more significant problems. Additionally, in order to guarantee the accuracy and reliability of data, the establishment of a Data and Security Monitoring Agency is important (3).

Poor culture, individual or organisational behaviour, leadership, procedures, or technology are the main causes of data integrity problems. Data integrity need to be properly applied to both paper and electronic documents, and it ought to be included into the quality management system. Workers have to get training on 21 CFR Part 11. To make sure that protocols are followed and audit trails are created, regular reviews and audits are necessary. Electronic solutions provide benefits over old paper-based methods in terms of better compliance with data integrity regulatory standards, in addition to being an efficient solution (system integration, data verification at both input and output, security) (4).

Furthermore, artificial intelligent are the newest technology used lately. The term “artificial intelligence” (AI) is used to describe how computers and other technologies may mimic human intellect and critical thinking. In 1956, John McCarthy was the first to define artificial intelligence (AI) as the study and development of methods for creating computers with minds. Alan Turing, in 1950, played a pivotal role in the establishment of contemporary computers and artificial intelligence. The “Turing test” was predicated on the notion that the hallmark of computer intelligence lies in its capacity to attain cognitive proficiency at par with humans (5).

During the 1980s and 1990s, there was a significant increase in the level of interest in artificial intelligence (AI). Various clinical contexts in healthcare have used artificial intelligent approaches, including fuzzy expert systems, Bayesian networks, artificial neural networks, and hybrid intelligent systems. In 2016, the healthcare industry received the largest portion of funding in AI research, surpassing other industries. AI techniques can detect and categorize hidden multivariate, nonlinear patterns in operational and performance data for plant engineers. Massive volumes of data are created by machines, sensors, controllers, and labor records nowadays. This data may be categorized: a) Data from ambient sensors, such as room temperature and humidity, b) Process data from sensors on machines or stations, such as machining and grinding coolant temperatures, power, and heat treat energy, c) production operation data stored in controller systems, e.g., timestamps or elapsed time of each component in each operation station, machine downtime, starvation/blockage, idle time, and shift scheduling, and d) Measure or verify product quality data, such as diameter, shape, and balance (6).

AI is used in the manufacturing business for Manufacturing System Optimization. This involves arranging production systems, which include machines, robots, conveyors, and supporting operations such as maintenance and material handling, in order to efficiently manufacture the intended output. Moreover, it may be used in the context of Human-Robot Collaboration (HRC), which refers to a situation where a specifically engineered robotic system and an operator collaborate on simultaneous activities inside a shared workspace, enabling both to carry out tasks simultaneously or together (7).

Moreover, AI may be used in Process Monitoring, Diagnostics, and Prognostics. The importance of maintenance and Prognostics and Health Management (PHM) systems in manufacturing lies in their ability to promptly detect the operational condition of individual machines or the production system. They offer a diagnosis of the underlying causes of anomalies and effectively prevent failures, resulting in minimal downtime. Furthermore, AI may be used in the supervision and regulation of industrial processes. This part is dedicated to the modeling and control of the process in order to get the required level of process quality. It assumes that any problems with the quality of the output are not caused by equipment failures. Furthermore, it explicitly investigates the impact of AI approaches in greatly improving the comprehension of the connection between materials and processes. This presents a promising chance for enhanced control, efficiency, quality, and productivity in the process (8).

Ultimately, the integration of AI in the industrial sector will provide certain hurdles to overcome, namely pertaining to data security and the maintenance of data integrity. This article will examine the use of artificial intelligence (AI) and data integrity in the manufacturing sector, investigating how the integration of these two elements leads to the attainment of data security resilience in line with ALCOA+ principles.

ALCOA+ Principles for AI Data Integrity

The pharmaceutical sector must rigorously adhere to data integrity and quality standards in order to minimise adverse effects and guarantee that medications meet the necessary quality standards and are safe for patients. ALCOA+ is a widely used data quality standard in the pharmaceutical sector. It consists of a set of guiding principles that are used to ensure data integrity. Noncompliance with ALCOA+ criteria, often identified during audit inspections, may lead to significant repercussions for pharmaceutical producers, including the imposition of penalties, escalation of expenses, and disruptions in production. The following are the fundamental principles of ALCOA+ (9).

AI plays a crucial role in improving data integrity by minimising human mistakes and enhancing data efficiency. Being processed. Furthermore, sophisticated algorithms and machine learning models, which are part of artificial intelligence (AI), have the ability to handle massive data sets with more accuracy than humans. As a result, they enhance the quality of data analysis. Contrarily, the effectiveness of AI systems depends on the quality of the data they get and the way they are built (Figure 1). Therefore, there is a fear that if the input data is incorrect or biassed, AI would magnify these problems. This viewpoint emphasises the moral obligation in managing data and the need of clearness in AI algorithms to guarantee the reliability of data. Consequently, effective data management necessitates addressing both the concerns of data integrity and data security. This underscores the need for an integrated tool capable of functioning collaboratively to simultaneously ensure the integrity and security of the data (10).

Figure 1. The principles of ALCOA+ (Attributable, Legible, Contemporaneous, Original, Accurate, Complete, Consistent, Enduring, Available), alongside considerations of AI ethics, regulatory compliance, data integrity certainty, and personal data protection, exert a significant impact on the field of artificial intelligence. These factors collectively contribute to shaping and establishing ethical and compliant practices within AI activities, thereby fostering the development of a robust and responsible AI system.

Oladoyinbo conducted a study investigating the associations between regulatory compliance, ethical awareness, professional training, and experience in the field of artificial intelligence (AI) with the effectiveness of AI implementation and the integrity of data.” The study results indicate a correlation between regulatory compliance in AI activities and the perceived efficacy in AI implementation. Higher levels of regulatory compliance in AI activities are strongly correlated with the perceived effectiveness in AI implementation. Furthermore, a correlation exists between Data Integrity certainty and AI ethics. AI practitioners that possess a heightened understanding of AI ethics exhibit a stronger level of certainty in maintaining data integrity inside their projects. Moreover, research has shown a positive correlation between expertise in the domain of artificial intelligence and the ability of an AI system to maintain data integrity (11).

The safeguarding of personal data as stipulated by the Personal Data Protection (PDP) Act and other legislations governing information privacy constitutes a pivotal consideration in the realm of data privacy. The definitions and interpretations of personal information are dynamic, evolving in tandem with societal and legal norms. There are several reasons why it is crucial to implement Personal Data Protection (PDP) (12):

• The ongoing progress of human rights,

• The emergence of technological advancements contributing to cybercrimes and vulnerabilities in personal data security,

• The global competition for data control leading to illicit data mining and the improper utilization of public personal data.

The foundational identifiability principle plays a crucial role in maintaining the integrity of data associated with personal information, particularly within the challenging landscape of distinguishing genuine from fabricated information in terms of identifiability. To address privacy concerns effectively, due consideration must be given to the three key factors of information or data: collection, purpose, and use. Implementation of the principles outlined by the Organization for Economic Cooperation and Development (OECD) is instrumental in achieving this objective. Specifically, the enduring pillars of collection restriction, purpose articulation, and use restriction serve as the fundamental tenets of information privacy derived from the OECD framework (13).

ALCOA+ Principles The Industry’s Struggles and Challenges in the Adoption of ALCOA+ Principles

Data Integrity, as defined by ISO/IEC 2382:2015, involves the maintenance of accuracy and consistency despite any alterations. In the context of pharmaceutical companies, Data Integrity assumes paramount significance, influencing critical facets like drug development, clinical trials, manufacturing, and adherence to regulatory standards. Uncompromised Data Integrity establishes confidence in the quality, efficacy, and safety of pharmaceuticals. Adhering to robust principles like ALCOA+ and complying with regulatory requirements such as FDA’s 21CFR11 and EMA’s EudraLex Annex 11, supported by thorough validation processes, ensures that Data Integrity remains central throughout the entire data lifecycle (14).

The lack of synergy and effective data management between companies poses challenges in standardizing protocols and procedures to ensure DI, irrespective of existing legislation and guidance (Figure 2). Moreover, protocols supporting DI in parent companies may not be adopted by their subsidiary counterparts. Failure to prevent DI violations could result in the release of substandard medicinal products into the market, posing risks of harm, potential fatalities, and, in the case of vaccines and biosimilars, eroding public confidence (15).

Figure 2. Indicators encompassing a lack of synergy and effective data management, economic downturns, inadequate involvement and resource allocation, as well as unmet key performance indicators (KPIs), can give rise to data integrity issues. These issues may manifest in various forms, such as inappropriate privilege levels and permissions, reintegrating or reprocessing inappropriately through manual integration, deletion or destruction of original Good Manufacturing Practice (GMP) records, discrepancies in audit trails, engagement in unofficial trial injections or repeat testing, maintenance of records that are not contemporaneous, instances of backdating or pre-dating, falsification of data, use of common shared passwords for log-in, reliance on unofficial and uncontrolled spreadsheets and notebooks, insufficient document control, utilization of non-validated methods, and delays or refusals in providing inspection access to original data.

The FDA reported an increase in data integrity warning letters from 2005 to 2017, averaging 12 warning letters per year. The most frequently observed data integrity issue was Incomplete or Missing Records, with a total of 107 reports, while Overwriting Results had the least occurrences, with only 7 reports. Additionally, there were several other violations of data integrity, including: Privilege levels, permissions inappropriate, Reintegrate, reprocess in appropriate manual integration, Deleting/destroying original GMP records, Audit trails, Unofficial, trial injections/ repeat testing, Records not contemporaneous, backdating, pre-dating, Falsified data, Back-up/restore, Log-in common shared password ETC, Unofficial, uncontrolled spreadsheets, notebooks, Inadequate document control, Method not validated, and Delay or refuse inspection, access to original data. Failure to uphold data integrity within an industry can lead to business losses, damage to reputation, regulatory scrutiny, competitive disadvantages, and diversion of resources to remediation efforts, ultimately resulting in an increase in attrition rates (16, 17).

Pharmaceutical companies frequently face pressure to enhance their key performance indicators (KPIs), particularly during economic downturns. This pressure has been associated with instances of data falsification aimed at reducing the rejection of manufactured batches. Some companies resort to practices such as deleting non-compliant records or generating records without legitimately conducting requisite tests to expedite regulatory approval. The lack of support from senior management, stemming from inadequate involvement and resource allocation, can exacerbate these situations. Additionally, employees may be apprehensive about potential job losses due to unmet KPIs, prompting them to release products without adhering to internal protocols requiring approval from authorized personnel or to manipulate records when granted access to the database. Notably, in systems involving manual data transfer to the company database using hybrid computerized systems, transcription errors may occur, resulting in inaccurate data records (18, 19).

Navigating the Challenges within the Realm of AI-Powered Digital Transformation and Its Application

Securing AI-enabled devices is of paramount importance in addressing social, economic, and environmental concerns. A significant number of recently developed AI models are susceptible to sophisticated hacking techniques. As a response, research in adversarial AI has intensified to design machine and deep learning models capable of withstanding diverse cyber threats. These embedded machine learning systems are now exposed to a novel class of risks termed artificial intelligence (AI) assaults. Consequently, they have become attractive targets for cyberattacks, posing a threat to the safety and security of larger interconnected systems. Contemporary AI assaults extend beyond mere coding errors and faults; rather, they stem from inherent flaws or limitations within the systems. Sangwan reports several types of failures in AI systems, including (20) :

• Exploitation of the requirement for enhanced resources for self-upgradation, which adversaries can manipulate.

• Implementation of malicious objectives that render AI systems unfriendly.

• Flaws in user-friendly features that compromise the integrity of the system.

• Utilization of diverse techniques to liberate different stages of AI from the constraints of actions, thereby exposing AI systems to potential adversaries.

In the implementation of AI in the industrial domain, careful attention must be given to the security of industry-related data. It is disconcerting that AI technologies are being widely employed despite their susceptibility to hostile attacks. Manipulation of input data with malicious intent has the potential to induce incorrect classifications in both Machine Learning (ML) and Deep Learning models. The context in which an algorithm operates significantly influences the multitude of adversarial attacks that may be launched against it. Although attacks against AI systems share similarities, the methods of exploitation vary depending on the specific algorithm employed (21).

Presently, manufacturing procedures undergo continuous modifications to proactively address customer needs and generate a diverse range of products. Numerous additional factors, such as machine malfunctions, alterations in orders, and unpredictable delays in work deliveries, can exert an impact on the production process within contemporary industrial settings. Several challenges in scheduling processes encompass (22) :

a) Single-machine scheduling. SMSP allocates jobs on a single machine to optimize an objective function.

b) The flow-shop scheduling problem. Tasks must be scheduled on computers in an FSSP. As the objects to be created must follow a specific sequence of tasks, each task will have a priority restriction on others. This sort of challenge has unidirectional material and information flow since all planned products must follow the same production sequence.

c) The job-shop scheduling problem. A collection of things must be processed on a set of machines in a JSSP, like an FSSP. Here, unlike an FSSP, objects may have different production sequences, therefore material flow is multi-directional.

d) Open-shop scheduling problem. An OSSP also has a set of components to process on a set of machines, but there are no priority limitations.

e) Parallel machine scheduling problem. A PMSP schedules tasks to run on many computers concurrently. How to assign work to machines and in what sequence is the main goal. An identical PMSP has all machines with the same processing speed and capabilities, whereas a uniform PMSP has machines within the same class with the same speed. Unrelated PMSPs have machines with different processing speeds.

Furthermore, AI in industry can also be used in the field of forecasting. It predicts future demand in order to strategize supply chain activities and operations with the goal of reducing shipping times, stocks, and operational expenditures. Forecasting has the capacity to anticipate future market demand; nevertheless, there are impediments within the forecasting process. These challenges encompass the identification of biases, limited support for and utilization of statistical analysis, a performance assessment exclusively centered on market forecasting, the manual input of data into the forecasting system, restricted access to performance metrics and visualization tools, as well as a deficiency in training and understanding pertaining to forecasting and its functionalities (Figure 3). AI has the capability to examine data and predict future demand, enhance the efficiency of logistics and transportation routes, and detect areas of congestion in the supply chain. Moreover, artificial intelligence (AI) contributes to enhancing the level of visibility in the supply chain, a crucial prerequisite for achieving a prosperous supply chain (23, 24). There are some AI techniques used for demand forecasting include Long Short-Term Memory (LSTM), Extreme Learning Machine (ELM), Adaptive-Network-Based Fuzzy Inference System (ANFIS), Machine learning (ML), traditional methode, Genetic-Algorithm, Neural-Network, Data-Mining Approach for Time-Series Models (GANNATS), and Dynamic Artificial Neural Networks (ANN) [25].

Figure 3. The integration of artificial intelligence with the establishment of robust data integrity emerges as a crucial solution for the digitization of industries. Various challenges in forecasting include the recognition of biases in the forecasting process, limited support and utilization of statistical analysis, performance evaluation exclusively focused on market forecasting, manual entry of data into the forecasting system, constrained availability of performance metrics and visualization tools, as well as inadequate training and knowledge regarding forecasting and its functionalities. Additionally, several challenges in scheduling encompass the open-shop scheduling problem, the job-shop scheduling problem, the flow-shop scheduling problem, and the single-machine scheduling problem.

2. CONCLUSION

ALCOA+ principles serve as key indicators for effective data management in various industries. The widespread use of artificial intelligence across domains brings both opportunities and challenges, particularly in data security and ethical considerations. The integration of AI with robust data integrity practices is identified as a pivotal solution for industry digitization. Furthermore, the challenges in forecasting and scheduling highlight the complexity in managing data processes. These challenges underscore the need for improved statistical analysis, comprehensive training, and enhanced tools for accurate forecasting and scheduling. The principles of ALCOA+, alongside considerations of AI ethics and personal data protection, play a significant role in shaping ethical practices within AI activities. The identified challenges in data integrity and the proposed solution of integrating AI with robust data practices collectively contribute to the ongoing discourse on responsible and ethical practices in the evolving digital era.

Author’s contributiion:

Both authors were involved in all steps of preparation this article, including final proofreading.

Conflicts of Interest:

The authors declare no conflict of interest.

Financial support and sponsorship:

This research received no external funding.