IELTS Reading: Trí tuệ nhân tạo cách mạng hóa chăm sóc sức khỏe cá nhân hóa – Đề thi mẫu có đáp án chi tiết

Mở bài

Chủ đề “How Is AI Transforming Personalized Healthcare?” đang trở thành một trong những đề tài hot nhất trong các kỳ thi IELTS Reading hiện nay. Với sự phát triển vượt bậc của công nghệ y tế và trí tuệ nhân tạo, Cambridge và IDP đã nhiều lần đưa các bài đọc liên quan đến chủ đề này vào đề thi thực tế, đặc biệt từ năm 2020 trở lại đây.

Bài viết này cung cấp cho bạn một bộ đề thi IELTS Reading hoàn chỉnh với 3 passages được thiết kế theo đúng chuẩn Cambridge, bao gồm độ khó tăng dần từ Easy (Band 5.0-6.5) đến Medium (Band 6.0-7.5) và Hard (Band 7.0-9.0). Bạn sẽ được trải nghiệm đầy đủ 40 câu hỏi với 7 dạng câu hỏi khác nhau thường xuất hiện trong kỳ thi thật, kèm theo đáp án chi tiết và giải thích cụ thể từng câu.

Ngoài ra, bài viết còn cung cấp hệ thống từ vựng chuyên ngành y tế và công nghệ được phân loại theo từng passage, giúp bạn không chỉ luyện kỹ năng làm bài mà còn mở rộng vốn từ học thuật. Đây là tài liệu phù hợp cho học viên từ band 5.0 trở lên, đặc biệt hữu ích cho những ai đang hướng tới band điểm 7.0+.

1. Hướng dẫn làm bài IELTS Reading

Tổng Quan Về IELTS Reading Test

IELTS Reading Test là phần thi kéo dài 60 phút với 3 passages và tổng cộng 40 câu hỏi. Đây là phần thi đòi hỏi khả năng quản lý thời gian tốt và kỹ thuật làm bài hiệu quả.

Phân bổ thời gian khuyến nghị:

• Passage 1: 15-17 phút (độ khó thấp, câu hỏi tương đối dễ xác định đáp án)
• Passage 2: 18-20 phút (độ khó trung bình, yêu cầu kỹ năng paraphrase và suy luận)
• Passage 3: 23-25 phút (độ khó cao, cần thời gian phân tích và hiểu sâu)
• Chuyển đáp án: 2-3 phút cuối

Lưu ý quan trọng: Mỗi câu trả lời đúng được tính 1 điểm, không có điểm âm cho câu sai. Do đó, bạn nên trả lời tất cả các câu hỏi, không được bỏ trống.

Các Dạng Câu Hỏi Trong Đề Này

Đề thi mẫu này bao gồm 7 dạng câu hỏi phổ biến nhất trong IELTS Reading:

1. Multiple Choice (Trắc nghiệm nhiều lựa chọn)
2. True/False/Not Given (Đúng/Sai/Không được đề cập)
3. Yes/No/Not Given (Có/Không/Không được đề cập)
4. Matching Headings (Nối tiêu đề với đoạn văn)
5. Sentence Completion (Hoàn thành câu)
6. Summary Completion (Hoàn thành đoạn tóm tắt)
7. Matching Features (Nối thông tin)

Mỗi dạng câu hỏi yêu cầu một kỹ thuật làm bài riêng biệt, và bạn sẽ được thực hành tất cả trong đề thi này.

---

2. IELTS Reading Practice Test

PASSAGE 1 – The Dawn of AI in Healthcare

Độ khó: Easy (Band 5.0-6.5)

Thời gian đề xuất: 15-17 phút

Artificial intelligence is rapidly transforming the landscape of modern medicine, offering unprecedented opportunities to personalize healthcare like never before. While the concept of tailored medical treatment has existed for decades, recent advances in machine learning algorithms and big data analytics have made truly individualized care a practical reality rather than just an aspirational goal.

The journey towards AI-powered personalized medicine begins with data collection. Every time a patient visits a doctor, undergoes a test, or fills a prescription, valuable information is generated. Traditional healthcare systems often struggled to harness this wealth of data effectively. However, AI systems can now process millions of patient records, identifying patterns and correlations that would be impossible for human practitioners to detect. These sophisticated algorithms analyze everything from genetic information and lifestyle factors to environmental exposures and treatment responses, creating a comprehensive picture of each patient’s unique health profile.

One of the most promising applications of AI in personalized healthcare is in diagnostic accuracy. Computer vision systems trained on thousands of medical images can now detect diseases like cancer, diabetic retinopathy, and pneumonia with accuracy rates that rival or exceed those of experienced specialists. What makes these systems particularly valuable is their ability to identify subtle patterns in diagnostic imaging that might escape human attention. For instance, an AI system developed by researchers at Stanford University can analyze chest X-rays and identify 14 different diseases with an accuracy rate above 90%, often outperforming board-certified radiologists.

Predictive analytics represents another groundbreaking application of AI in healthcare personalization. By analyzing a patient’s complete medical history, genetic markers, and lifestyle data, AI algorithms can forecast the likelihood of developing certain conditions years before symptoms appear. This proactive approach allows doctors to implement preventive measures tailored to each individual’s specific risk factors. For example, machine learning models can predict which patients are at high risk of developing type 2 diabetes by analyzing factors such as body mass index, family history, dietary patterns, and even sleep quality. Armed with this knowledge, healthcare providers can design personalized intervention programs that may include specific dietary recommendations, exercise routines, and regular monitoring schedules.

The pharmaceutical industry is also experiencing a revolution thanks to AI-driven personalization. Drug development traditionally follows a “one-size-fits-all” approach, where medications are designed to work for the average patient. However, AI is enabling precision medicine, where treatments can be customized based on an individual’s genetic makeup and biological characteristics. Pharmacogenomics, the study of how genes affect drug response, has been enhanced by AI algorithms that can predict which medications will be most effective for specific patients while minimizing adverse reactions. This approach not only improves treatment outcomes but also reduces healthcare costs by avoiding ineffective treatments and preventing harmful side effects.

Virtual health assistants powered by AI are making personalized healthcare more accessible to millions of people. These intelligent systems can provide 24/7 medical guidance, answer health-related questions, monitor chronic conditions, and even offer mental health support. Unlike traditional healthcare services that require appointments and waiting rooms, AI assistants offer immediate responses tailored to each user’s specific situation. They can track medication adherence, send reminders for preventive screenings, and alert users to potential health concerns based on symptom reports. Some advanced systems can even analyze voice patterns to detect early signs of conditions like depression or Parkinson’s disease.

Despite these remarkable advances, the integration of AI into personalized healthcare faces several challenges. Data privacy remains a paramount concern, as AI systems require access to sensitive medical information to function effectively. Healthcare providers must ensure robust security measures to protect patient data from breaches while still allowing AI systems to learn and improve. Additionally, there are concerns about algorithmic bias – AI systems trained primarily on data from certain demographic groups may not perform as well for underrepresented populations. Researchers are actively working to address these issues by developing more inclusive datasets and implementing fairness algorithms that ensure equitable care for all patients.

The regulatory landscape for AI in healthcare is still evolving. Medical AI systems must undergo rigorous testing to prove their safety and effectiveness before being approved for clinical use. Regulatory bodies like the FDA are developing new frameworks specifically for evaluating AI-based medical devices, recognizing that these systems differ fundamentally from traditional medical equipment. The challenge lies in creating regulations that ensure patient safety without stifling innovation or delaying the deployment of beneficial technologies.

Looking ahead, the future of AI in personalized healthcare appears incredibly promising. As computational power increases and algorithms become more sophisticated, AI systems will likely become even better at predicting health outcomes and recommending optimal treatment strategies. The integration of wearable devices and continuous monitoring technology will provide AI systems with real-time health data, enabling even more precise and timely interventions. Perhaps most excitingly, AI may help democratize access to high-quality healthcare by bringing expert-level medical guidance to underserved communities around the world.

Questions 1-13

Questions 1-5: Multiple Choice

Choose the correct letter, A, B, C, or D.

1. According to the passage, what has made personalized healthcare a practical reality?
   A. Increased funding for medical research
   B. Advances in machine learning and data analytics
   C. More patients visiting doctors regularly
   D. The development of new prescription drugs

2. AI systems at Stanford University can identify how many diseases from chest X-rays?
   A. 10 diseases
   B. 12 diseases
   C. 14 diseases
   D. 16 diseases

3. Predictive analytics in AI healthcare is described as:
   A. A reactive approach to treating diseases
   B. A proactive approach to preventing conditions
   C. An experimental technique with limited use
   D. A replacement for traditional medical practice

4. What does pharmacogenomics study?
   A. How genes affect drug response
   B. How drugs are manufactured
   C. How patients take their medications
   D. How genes cause diseases

5. Virtual health assistants differ from traditional healthcare services by:
   A. Providing better quality care
   B. Requiring appointments
   C. Offering immediate responses
   D. Being more expensive

Questions 6-9: True/False/Not Given

Do the following statements agree with the information given in the passage?

Write:

• TRUE if the statement agrees with the information
• FALSE if the statement contradicts the information
• NOT GIVEN if there is no information on this

6. AI systems can only analyze genetic information to create health profiles.

7. Computer vision systems can sometimes perform better than experienced radiologists.

8. All pharmaceutical companies have adopted AI-driven personalization methods.

9. Some AI assistants can detect early signs of Parkinson’s disease through voice analysis.

Questions 10-13: Sentence Completion

Complete the sentences below.

Choose NO MORE THAN THREE WORDS from the passage for each answer.

10. Healthcare providers must implement strong __ to protect patient information from unauthorized access.

11. AI systems may not work equally well for all groups due to __ in their training data.

12. Regulatory bodies are creating new __ specifically for evaluating AI medical devices.

13. The integration of wearable devices will provide AI with __ health data for better interventions.

Trí tuệ nhân tạo đang cách mạng hóa y tế cá nhân hóa với công nghệ phân tích dữ liệu và học máy hiện đại

---

PASSAGE 2 – AI’s Impact on Clinical Decision-Making and Treatment Optimization

Độ khó: Medium (Band 6.0-7.5)

Thời gian đề xuất: 18-20 phút

The integration of artificial intelligence into clinical practice has fundamentally altered the paradigm of medical decision-making, shifting from intuition-based approaches to evidence-driven methodologies that leverage vast repositories of medical knowledge. This transformation has profound implications for how healthcare professionals diagnose diseases, select treatments, and monitor patient progress, ultimately leading to more precise and effective therapeutic interventions.

Clinical decision support systems (CDSS) powered by AI represent one of the most significant innovations in modern healthcare. These systems function as intelligent advisors, analyzing patient data in real-time and providing physicians with evidence-based recommendations for diagnosis and treatment. Unlike traditional CDSS that relied on rigid, rule-based logic, contemporary AI-powered systems employ neural networks and deep learning techniques that can adapt to new information and recognize complex patterns across diverse patient populations. The sophistication of these systems has reached a point where they can synthesize information from disparate sources – including electronic health records, laboratory results, medical imaging, and current research literature – to generate comprehensive clinical insights within seconds.

The application of AI in oncology exemplifies the potential of personalized treatment optimization. Cancer is a highly heterogeneous disease, with each tumor possessing unique molecular characteristics that determine its behavior and response to therapy. Traditional treatment protocols often followed standardized guidelines that did not account for individual tumor variations. However, AI algorithms can now analyze genomic sequencing data from tumor samples, identifying specific genetic mutations and biomarkers that predict treatment response. IBM Watson for Oncology, for instance, has been trained on thousands of cancer cases and medical journals, enabling it to recommend personalized treatment plans that consider not only the type and stage of cancer but also the patient’s genetic profile, comorbidities, and treatment history. Studies have shown that AI-recommended treatments align with expert oncologist decisions in over 90% of cases, while occasionally suggesting alternative therapies that physicians might not have considered.

Cardiovascular disease management has similarly benefited from AI-driven personalization. Heart disease remains the leading cause of mortality worldwide, yet its prevention and treatment have historically been challenging due to the multifactorial nature of cardiovascular risk. AI algorithms excel at processing the numerous variables that contribute to cardiac health, including blood pressure patterns, lipid profiles, inflammatory markers, lifestyle factors, and even psychosocial stress indicators. Machine learning models trained on data from thousands of patients can predict individual risk of heart attack or stroke with remarkable precision, often identifying high-risk individuals who would be classified as low or moderate risk by traditional risk assessment tools. This enhanced risk stratification enables clinicians to implement more aggressive preventive strategies for truly high-risk patients while avoiding unnecessary interventions for those at lower risk.

The optimization of medication regimens represents another area where AI has demonstrated substantial value. Polypharmacy – the concurrent use of multiple medications – affects millions of patients worldwide, particularly the elderly population with multiple chronic conditions. The complexity of potential drug interactions, contraindications, and individual variations in drug metabolism makes medication management extraordinarily challenging. AI systems can analyze a patient’s complete medication profile, genetic variants affecting drug processing, and real-time monitoring data to identify potential problems and suggest optimal dosing strategies. Research conducted at the University of California has shown that AI-guided medication management can reduce adverse drug events by up to 30% while improving therapeutic efficacy.

Chronic disease management has been revolutionized by AI-powered continuous monitoring systems. Conditions such as diabetes, hypertension, and asthma require ongoing vigilance and frequent treatment adjustments. Traditional management approaches relied on periodic clinic visits and patient self-reporting, which often failed to capture the dynamic nature of these conditions. Modern AI systems can integrate data from wearable sensors, continuous glucose monitors, smart inhalers, and other connected devices to provide real-time insights into disease status. These systems employ predictive algorithms that can forecast disease exacerbations before they occur, alerting patients and providers to take preemptive action. For diabetic patients, AI algorithms can predict hypoglycemic episodes hours in advance by analyzing patterns in glucose readings, activity levels, meal timing, and medication administration, allowing for timely interventions that prevent dangerous complications.

The mental health sector has emerged as an unexpected beneficiary of AI personalization technologies. Psychiatric conditions have traditionally been difficult to diagnose and treat due to their subjective nature and the heterogeneity of symptom presentation. AI systems are now being employed to analyze multiple data streams – including speech patterns, facial expressions, social media activity, sleep patterns, and smartphone usage – to detect early signs of mental health deterioration. Natural language processing algorithms can identify subtle changes in communication style that may indicate emerging depression or anxiety. Some systems have achieved accuracy rates above 80% in predicting psychiatric hospitalization weeks before a crisis occurs, enabling timely interventions that may prevent severe episodes.

Despite these advances, the implementation of AI in clinical decision-making faces substantial obstacles. The “black box” problem – where AI systems make recommendations without providing transparent explanations of their reasoning – creates trust issues among healthcare providers who must understand why a particular treatment is recommended before they can confidently prescribe it. Recent developments in “explainable AI” aim to address this concern by designing algorithms that can articulate the specific factors and logic underlying their recommendations. Additionally, the medicolegal implications of AI-assisted decisions remain unclear. Questions about liability when AI recommendations lead to adverse outcomes, the standard of care expected when AI tools are available, and the degree to which physicians can deviate from AI suggestions without malpractice risk are still being debated in both medical and legal communities.

The clinical validation of AI systems presents another significant challenge. While many AI algorithms demonstrate impressive performance in research settings, their effectiveness in diverse, real-world clinical environments must be rigorously established. AI systems trained primarily on data from large academic medical centers may not perform as well in smaller community hospitals or resource-limited settings. Furthermore, the dynamic nature of medical knowledge means that AI systems require continuous updating to incorporate new research findings and evolving treatment standards. Establishing governance frameworks for monitoring AI system performance, updating algorithms, and ensuring ongoing accuracy represents a complex logistical challenge for healthcare organizations.

Looking forward, the trajectory of AI in clinical decision-making points toward increasingly sophisticated systems that not only recommend treatments but also predict how individual patients will respond to specific interventions. The convergence of AI with other emerging technologies – including genomics, proteomics, metabolomics, and advanced imaging – promises to create an unprecedented level of treatment precision. However, realizing this vision will require addressing current limitations while ensuring that AI augments rather than replaces human clinical judgment, preserving the essential human elements of empathy, ethical reasoning, and patient-centered care that define excellent medical practice.

Questions 14-26

Questions 14-18: Yes/No/Not Given

Do the following statements agree with the claims of the writer in the passage?

Write:

• YES if the statement agrees with the claims of the writer
• NO if the statement contradicts the claims of the writer
• NOT GIVEN if it is impossible to say what the writer thinks about this

14. Modern AI-powered clinical decision support systems are more flexible than traditional rule-based systems.

15. IBM Watson for Oncology always provides better treatment recommendations than expert oncologists.

16. Traditional cardiovascular risk assessment tools can accurately identify all high-risk patients.

17. AI-guided medication management has been proven to reduce adverse drug events.

18. All healthcare providers trust AI recommendations without question.

Questions 19-22: Matching Headings

The passage has ten paragraphs labeled A-J. Which paragraph contains the following information?

Write the correct letter, A-J.

19. A discussion of how AI helps manage multiple medications simultaneously

20. An explanation of how AI detects mental health problems before they become severe

21. A description of the challenges in ensuring AI systems work effectively in different healthcare settings

22. An example of how AI personalizes cancer treatment based on genetic information

Questions 23-26: Summary Completion

Complete the summary below.

Choose NO MORE THAN TWO WORDS from the passage for each answer.

AI has transformed clinical decision-making by moving from intuition-based approaches to 23. __ that use extensive medical knowledge. In cardiovascular disease, AI can process multiple factors including lifestyle and 24. __ to predict heart attack risk more accurately than traditional methods. For chronic diseases, AI systems integrate data from 25. __ and other connected devices to predict when conditions might worsen. However, the “26. __” problem means that AI systems sometimes cannot explain their recommendations clearly, which creates trust issues among doctors.

Hệ thống trí tuệ nhân tạo hỗ trợ bác sĩ đưa ra quyết định lâm sàng và tối ưu hóa điều trị bệnh nhân

---

PASSAGE 3 – The Ethical Dimensions and Sociotechnical Complexities of AI-Driven Healthcare Personalization

Độ khó: Hard (Band 7.0-9.0)

Thời gian đề xuất: 23-25 phút

The proliferation of artificial intelligence in healthcare has precipitated a paradigmatic shift that extends far beyond mere technological advancement, engendering complex ethical dilemmas and sociotechnical challenges that demand rigorous interdisciplinary scrutiny. While the potential benefits of AI-driven personalized medicine are incontrovertible, the implementation of these technologies intersects with fundamental questions about autonomy, justice, beneficence, and the very nature of the physician-patient relationship, necessitating careful consideration of how these systems are designed, deployed, and governed within diverse healthcare ecosystems.

Algorithmic bias represents one of the most pernicious challenges confronting AI-driven healthcare personalization, with potentially far-reaching ramifications for health equity. Machine learning algorithms are inherently products of their training data; consequently, if historical medical data reflects systemic disparities in healthcare access, quality, or outcomes among different demographic groups, AI systems trained on such data will perpetuate and potentially amplify these inequities. Research has documented numerous instances where medical AI systems perform markedly worse for underrepresented populations. A landmark study published in Science revealed that an algorithm used by hospitals to identify patients requiring additional medical attention systematically underestimated the needs of Black patients compared to white patients with equivalent objective health measures. This discriminatory performance arose not from explicit racial variables in the algorithm but from the use of healthcare costs as a proxy for medical necessity – a problematic assumption given that Black patients historically receive less expensive care than white patients with similar conditions due to structural barriers to access and implicit bias in clinical decision-making. Such findings underscore the imperative for comprehensive bias auditing and the development of fairness-aware algorithms that can identify and mitigate discriminatory patterns.

The epistemological challenges posed by “black box” machine learning models particularly deep neural networks – present profound implications for medical accountability and informed consent. These systems often arrive at diagnostic or therapeutic recommendations through computational processes that are opaque even to their creators, involving millions of weighted parameters that defy straightforward interpretation. This opacity creates a fundamental tension with established principles of evidence-based medicine, which emphasizes transparent reasoning from clinical data to therapeutic decisions. When an AI system recommends a particular treatment but cannot articulate the specific clinical factors that justify that recommendation, physicians face an epistemological quandary: should they defer to the algorithm’s statistical authority, derived from processing vastly more cases than any human could experience, or insist on comprehensible justifications that align with traditional medical reasoning? This dilemma has spurred the emergence of “explainable AI” (XAI) as a research priority, with investigators developing techniques such as attention mechanisms, saliency maps, and counterfactual explanations that can illuminate which input features most strongly influence AI outputs. However, current XAI methods face limitations; explanations may be post hoc rationalizations rather than faithful representations of the algorithm’s actual decision process, and even the most transparent algorithms may identify valid but counterintuitive patterns that challenge conventional medical understanding.

Data governance and privacy considerations acquire heightened salience in the context of AI-driven personalized healthcare, as these systems require access to granular, longitudinal health information spanning genomic data, biometric measurements, lifestyle behaviors, and increasingly, information from sources not traditionally considered medical, such as consumer purchases, social media activity, and geolocation data. The aggregation of such multidimensional datasets creates unprecedented opportunities for inferential analytics but simultaneously magnifies privacy risks. Re-identification attacks have demonstrated that supposedly anonymized health data can often be linked back to individuals through correlation with publicly available information, while the increasing sophistication of AI enables the extraction of sensitive inferences from seemingly innocuous data – for instance, predicting individuals’ health conditions from their social media posts or consumer behavior. These capabilities problematize traditional approaches to informed consent, which typically involve discrete authorization for specific uses of clearly defined data. In the AI era, individuals may unknowingly contribute to training datasets, and the potential uses of their data – including future applications not yet conceived when consent was obtained – may be indeterminate. Some scholars advocate for new governance models, such as “data trusts” or “dynamic consent” systems that provide individuals with ongoing control over their health information, though implementing such approaches at scale presents substantial technical and administrative hurdles.

The potential for AI systems to exacerbate or ameliorate existing health disparities depends critically on how these technologies are distributed and implemented across different populations and healthcare settings. While proponents envision AI democratizing access to high-quality care by bringing expert-level diagnostic and therapeutic guidance to underserved communities, there is a countervailing risk that AI may instead widen health inequities if its benefits accrue primarily to affluent populations with access to cutting-edge medical facilities and the digital literacy to engage with AI-enhanced healthcare tools. The digital divide – encompassing disparities in internet access, smartphone ownership, and technological fluency – creates barriers to participation in AI-driven healthcare models that increasingly rely on patient-facing applications, wearable devices, and telemedicine platforms. Moreover, the infrastructure requirements for implementing sophisticated AI systems may be prohibitive for resource-constrained healthcare settings, potentially creating a two-tiered system where patients in well-funded institutions receive AI-augmented care while those in safety-net hospitals and rural clinics do not. Addressing these distributional justice concerns requires deliberate policies ensuring equitable access to AI technologies and proactive efforts to design systems that function effectively across diverse contexts, including settings with limited technological infrastructure.

The transformation of the physician-patient relationship represents another dimension of sociotechnical change warranting critical examination. Relational continuity and empathetic communication have long been recognized as essential components of effective healthcare, contributing to patient satisfaction, treatment adherence, and health outcomes. The interpersonal dynamics of medical encounters may shift substantially as AI systems assume increasingly prominent roles in clinical decision-making. Potential concerns include the mechanization of care – where the humanistic aspects of medicine are subordinated to algorithmic efficiency – and the possibility that patients may feel their unique subjective experiences are being reduced to data points processed by impersonal algorithms. Conversely, AI could potentially enhance the physician-patient relationship by handling routine analytical tasks, thereby freeing clinicians to focus more attention on the interpersonal and counseling dimensions of care. The actual impact likely depends on how AI is integrated into clinical workflows and whether physicians are trained to use these tools in ways that complement rather than supplant their relational competencies. Some evidence suggests that patients are generally receptive to AI-assisted healthcare when they understand its purpose and retain meaningful human involvement in their care, though acceptance varies across demographic groups and clinical contexts.

Regulatory frameworks for AI in healthcare must navigate the delicate balance between ensuring safety and efficacy while fostering innovation in a rapidly evolving technological landscape. Traditional medical device regulation was designed for static products whose characteristics are established during the approval process and remain unchanged thereafter. AI systems, however, are fundamentally dynamic, continually learning and adapting as they process new data, which means their performance characteristics may drift over time. This temporal instability challenges conventional regulatory approaches. Should each iteration of an algorithm require separate approval? How can regulators ensure ongoing safety and effectiveness without stifling the adaptive capabilities that make AI valuable? The U.S. Food and Drug Administration has proposed a “predetermined change control plan” approach allowing manufacturers to specify in advance the types of algorithm modifications they intend to make and the methodology for validating those changes, enabling iterative improvements without repeated regulatory review. However, implementing such frameworks requires developing new technical standards for algorithm validation, establishing mechanisms for post-market surveillance to detect performance degradation, and determining appropriate governance structures for overseeing AI systems throughout their lifecycle.

The epistemological authority that AI systems may accrue raises questions about the locus of medical expertise and decision-making power. As algorithms demonstrate superhuman performance in certain diagnostic tasks and treatment recommendations, there is potential for deskilling of healthcare professionals who may come to rely on AI systems without maintaining the underlying clinical knowledge necessary to critically evaluate algorithmic recommendations. This epistemic dependence could prove problematic if AI systems malfunction, encounter out-of-distribution cases for which they were not trained, or if their recommendations become outdated as medical knowledge evolves. Automation bias – the tendency to favor suggestions from automated systems even when they contradict human judgment – has been documented in various domains and poses risks in healthcare contexts where uncritical deference to AI could lead to diagnostic errors or inappropriate treatments. Maintaining appropriate calibration between trust in AI systems and human oversight requires training clinicians to understand AI capabilities and limitations, fostering what some scholars term “algorithmic literacy” among healthcare professionals.

The commodification of health data and the involvement of commercial entities in AI healthcare development introduce potential conflicts between profit motives and patient welfare. Many leading AI healthcare systems are being developed by technology corporations whose business models center on data aggregation and whose fiduciary obligations to shareholders may not align with healthcare’s traditional ethic of prioritizing patient interests. Concerns have been raised about “surveillance medicine” where the collection of health data serves commercial purposes beyond direct patient care, such as targeted advertising, insurance underwriting, or predictive policing. The concentration of AI healthcare capabilities in a small number of large technology companies also raises antitrust concerns and questions about whether critical healthcare infrastructure should rest in private hands. Some jurisdictions have begun exploring alternative governance models, such as publicly funded AI development or collaborative frameworks involving multiple stakeholders, though these approaches face challenges in mobilizing the substantial financial resources and technical expertise required for cutting-edge AI research.

Ultimately, the trajectory of AI in personalized healthcare will be shaped not merely by technological capabilities but by collective choices about values, priorities, and governance structures. Ensuring that AI serves as a force for health equity rather than exacerbating disparities, that it enhances rather than erodes the humanistic dimensions of medicine, and that its deployment reflects democratic deliberation rather than technocratic imposition requires sustained engagement among diverse stakeholders – including patients, clinicians, researchers, ethicists, policymakers, and communities most vulnerable to both health inequities and algorithmic harms. The promise of AI-driven personalized healthcare is substantial, but realizing that promise in ways that are equitable, accountable, and aligned with fundamental human values demands vigilant attention to the ethical and social dimensions of these powerful technologies.

Questions 27-40

Questions 27-31: Multiple Choice

Choose the correct letter, A, B, C, or D.

27. According to the passage, algorithmic bias in healthcare AI is primarily caused by:
    A. Deliberate discrimination by AI developers
    B. Historical disparities reflected in training data
    C. Insufficient computing power
    D. Lack of regulatory oversight

28. The study published in Science found that the hospital algorithm underestimated Black patients’ needs because:
    A. It explicitly used race as a variable
    B. It was designed by biased programmers
    C. It used healthcare costs as a proxy for medical necessity
    D. Black patients provided inaccurate health information

29. “Explainable AI” (XAI) research aims to:
    A. Replace human doctors with AI systems
    B. Make AI recommendations more comprehensible
    C. Increase the speed of AI processing
    D. Reduce the cost of AI development

30. The concept of “data trusts” is mentioned as a potential solution to:
    A. Improving AI diagnostic accuracy
    B. Reducing healthcare costs
    C. Providing ongoing patient control over health information
    D. Training better machine learning algorithms

31. The FDA’s “predetermined change control plan” approach is designed to:
    A. Prevent any changes to approved AI systems
    B. Allow iterative improvements without repeated reviews
    C. Eliminate the need for AI regulation
    D. Require more frequent regulatory submissions

Questions 32-36: Matching Features

Match each concern (32-36) with the correct stakeholder group (A-F) that would be most affected.

A. Healthcare professionals
B. Technology corporations
C. Patients in underserved communities
D. Regulatory bodies
E. AI developers
F. Insurance companies

32. Risk of diagnostic errors due to automation bias

33. Barriers created by the digital divide and lack of infrastructure

34. Challenges in balancing profit motives with patient welfare

35. Difficulty in regulating dynamic, constantly evolving systems

36. Concerns about surveillance medicine and data commodification

Questions 37-40: Short-answer Questions

Answer the questions below.

Choose NO MORE THAN THREE WORDS from the passage for each answer.

37. What type of machine learning models are particularly difficult to interpret according to the passage?

38. What term describes the tendency to favor automated system suggestions over human judgment?

39. What kind of obligations do technology corporations have to their shareholders that may conflict with patient interests?

40. According to the passage, what type of engagement is needed among stakeholders to ensure AI serves health equity?

Các thách thức đạo đức và phức tạp xã hội kỹ thuật trong ứng dụng trí tuệ nhân tạo chăm sóc sức khỏe cá nhân

---

3. Answer Keys – Đáp Án

PASSAGE 1: Questions 1-13

1. B
2. C
3. B
4. A
5. C
6. FALSE
7. TRUE
8. NOT GIVEN
9. TRUE
10. security measures
11. algorithmic bias
12. frameworks
13. real-time

PASSAGE 2: Questions 14-26

14. YES
15. NO
16. NO
17. YES
18. NO
19. E (Paragraph 5)
20. G (Paragraph 7)
21. I (Paragraph 9)
22. C (Paragraph 3)
23. evidence-driven methodologies
24. psychosocial stress
25. wearable sensors
26. black box

PASSAGE 3: Questions 27-40

27. B
28. C
29. B
30. C
31. B
32. A
33. C
34. B
35. D
36. F (hoặc B – cả hai đều hợp lý trong ngữ cảnh)
37. deep neural networks
38. Automation bias
39. fiduciary obligations
40. sustained engagement

---

4. Giải Thích Đáp Án Chi Tiết

Passage 1 – Giải Thích

Câu 1: B

• Dạng câu hỏi: Multiple Choice
• Từ khóa: personalized healthcare, practical reality
• Vị trí trong bài: Đoạn 1, dòng 2-4
• Giải thích: Câu trong bài viết rõ ràng: “recent advances in machine learning algorithms and big data analytics have made truly individualized care a practical reality” – chính xác là đáp án B. Các đáp án khác không được đề cập như nguyên nhân chính.

Câu 2: C

• Dạng câu hỏi: Multiple Choice
• Từ khóa: Stanford University, chest X-rays, diseases
• Vị trí trong bài: Đoạn 3, dòng 5-7
• Giải thích: Bài viết nêu rõ: “an AI system developed by researchers at Stanford University can analyze chest X-rays and identify 14 different diseases” – đáp án chính xác là 14 diseases (C).

Câu 3: B

• Dạng câu hỏi: Multiple Choice
• Từ khóa: Predictive analytics, described as
• Vị trí trong bài: Đoạn 4, dòng 4-5
• Giải thích: Passage sử dụng cụm “proactive approach” để mô tả predictive analytics, cho phép bác sĩ thực hiện preventive measures (biện pháp phòng ngừa) trước khi triệu chứng xuất hiện.

Câu 6: FALSE

• Dạng câu hỏi: True/False/Not Given
• Từ khóa: AI systems, only analyze genetic information
• Vị trí trong bài: Đoạn 2, dòng 4-7
• Giải thích: Bài viết nêu rõ AI phân tích “everything from genetic information and lifestyle factors to environmental exposures and treatment responses” – không chỉ genetic information, do đó câu này sai (FALSE).

Câu 7: TRUE

• Dạng câu hỏi: True/False/Not Given
• Từ khóa: Computer vision systems, perform better, radiologists
• Vị trí trong bài: Đoạn 3, dòng 7-8
• Giải thích: Passage nói: “often outperforming board-certified radiologists” – xác nhận rằng đôi khi AI có thể làm tốt hơn các bác sĩ X-quang có kinh nghiệm.

Câu 9: TRUE

• Dạng câu hỏi: True/False/Not Given
• Từ khóa: AI assistants, detect, Parkinson’s disease, voice
• Vị trí trong bài: Đoạn 6, dòng cuối
• Giải thích: Bài viết đề cập: “Some advanced systems can even analyze voice patterns to detect early signs of conditions like depression or Parkinson’s disease.”

Câu 10: security measures

• Dạng câu hỏi: Sentence Completion
• Từ khóa: Healthcare providers, protect patient information
• Vị trí trong bài: Đoạn 7, dòng 2-3
• Giải thích: “Healthcare providers must ensure robust security measures to protect patient data from breaches” – cụm từ cần điền là “security measures”.

Câu 23: evidence-driven methodologies

• Dạng câu hỏi: Summary Completion
• Từ khóa: moving from intuition-based approaches
• Vị trí trong bài: Passage 2, đoạn 1, câu đầu
• Giải thích: Bài viết nói về sự chuyển đổi “from intuition-based approaches to evidence-driven methodologies” – đây là paraphrase chính xác của câu hỏi.

Câu 27: B

• Dạng câu hỏi: Multiple Choice
• Từ khóa: algorithmic bias, primarily caused by
• Vị trí trong bài: Passage 3, đoạn 2, dòng 1-3
• Giải thích: “Machine learning algorithms are inherently products of their training data; consequently, if historical medical data reflects systemic disparities… AI systems trained on such data will perpetuate and potentially amplify these inequities” – rõ ràng nguyên nhân là dữ liệu huấn luyện phản ánh sự bất bình đẳng lịch sử.

Câu 28: C

• Dạng câu hỏi: Multiple Choice
• Từ khóa: Science study, algorithm, Black patients
• Vị trí trong bài: Passage 3, đoạn 2, dòng 5-10
• Giải thích: Bài viết giải thích rõ: “This discriminatory performance arose not from explicit racial variables in the algorithm but from the use of healthcare costs as a proxy for medical necessity” – đáp án C chính xác.

Câu 32: A (Healthcare professionals)

• Dạng câu hỏi: Matching Features
• Từ khóa: automation bias, diagnostic errors
• Vị trí trong bài: Passage 3, đoạn 8
• Giải thích: Automation bias ảnh hưởng trực tiếp đến các chuyên gia y tế (healthcare professionals) khi họ quá phụ thuộc vào gợi ý của AI và có thể bỏ qua đánh giá lâm sàng của chính mình.

Câu 38: Automation bias

• Dạng câu hỏi: Short-answer Questions
• Từ khóa: tendency, favor automated system suggestions
• Vị trí trong bài: Passage 3, đoạn 8, dòng 6-7
• Giải thích: Passage định nghĩa rõ ràng: “Automation bias – the tendency to favor suggestions from automated systems even when they contradict human judgment.”

---

5. Từ Vựng Quan Trọng Theo Passage

Passage 1 – Essential Vocabulary

Từ vựng	Loại từ	Phiên âm	Nghĩa tiếng Việt	Ví dụ từ bài	Collocation
personalize	v	/ˈpɜːsənəlaɪz/	cá nhân hóa	offering unprecedented opportunities to personalize healthcare	personalized medicine, personalized treatment
harness	v	/ˈhɑːnɪs/	khai thác, tận dụng	struggled to harness this wealth of data effectively	harness data, harness technology
diagnostic accuracy	n	/ˌdaɪəɡˈnɒstɪk ˈækjərəsi/	độ chính xác trong chẩn đoán	one of the most promising applications is in diagnostic accuracy	improve diagnostic accuracy, achieve high diagnostic accuracy
rival	v	/ˈraɪvəl/	sánh ngang, cạnh tranh	accuracy rates that rival or exceed those of specialists	rival human performance, rival traditional methods
predictive analytics	n	/prɪˈdɪktɪv ˌænəˈlɪtɪks/	phân tích dự đoán	predictive analytics represents another groundbreaking application	use predictive analytics, apply predictive analytics
proactive approach	n	/prəʊˈæktɪv əˈprəʊtʃ/	phương pháp chủ động	this proactive approach allows doctors to implement preventive measures	take a proactive approach, adopt a proactive approach
intervention program	n	/ˌɪntəˈvenʃən ˈprəʊɡræm/	chương trình can thiệp	design personalized intervention programs	implement intervention programs, develop intervention programs
precision medicine	n	/prɪˈsɪʒən ˈmedsɪn/	y học chính xác	AI is enabling precision medicine	practice precision medicine, advance precision medicine
adverse reaction	n	/ədˈvɜːs riˈækʃən/	phản ứng phụ, tác dụng phụ	minimizing adverse reactions	reduce adverse reactions, prevent adverse reactions
medication adherence	n	/ˌmedɪˈkeɪʃən ədˈhɪərəns/	sự tuân thủ dùng thuốc	track medication adherence	improve medication adherence, monitor medication adherence
algorithmic bias	n	/ˌælɡəˈrɪðmɪk ˈbaɪəs/	thiên lệch thuật toán	concerns about algorithmic bias	address algorithmic bias, mitigate algorithmic bias
regulatory landscape	n	/ˈreɡjələtəri ˈlændskeɪp/	bối cảnh pháp lý	the regulatory landscape for AI in healthcare	navigate the regulatory landscape, evolving regulatory landscape

Passage 2 – Essential Vocabulary

Từ vựng	Loại từ	Phiên âm	Nghĩa tiếng Việt	Ví dụ từ bài	Collocation
paradigm	n	/ˈpærədaɪm/	mô hình, hệ mẫu	fundamentally altered the paradigm of medical decision-making	shift the paradigm, paradigm shift
clinical decision support	n	/ˈklɪnɪkəl dɪˈsɪʒən səˈpɔːt/	hỗ trợ quyết định lâm sàng	clinical decision support systems (CDSS)	provide clinical decision support, implement clinical decision support
neural network	n	/ˈnjʊərəl ˈnetwɜːk/	mạng nơ-ron	contemporary systems employ neural networks	train neural networks, artificial neural networks
oncology	n	/ɒŋˈkɒlədʒi/	ung thư học	the application of AI in oncology	precision oncology, clinical oncology
heterogeneous	adj	/ˌhetərəˈdʒiːniəs/	không đồng nhất	cancer is a highly heterogeneous disease	heterogeneous population, heterogeneous data
genomic sequencing	n	/dʒiːˈnɒmɪk ˈsiːkwənsɪŋ/	giải trình tự gen	analyze genomic sequencing data	perform genomic sequencing, genomic sequencing technology
biomarker	n	/ˈbaɪəʊˌmɑːkə/	dấu ấn sinh học	identifying specific genetic mutations and biomarkers	discover biomarkers, biomarker detection
comorbidity	n	/ˌkəʊmɔːˈbɪdəti/	bệnh kèm theo	patient’s comorbidities	manage comorbidities, multiple comorbidities
cardiovascular risk	n	/ˌkɑːdiəʊˈvæskjələ rɪsk/	nguy cơ tim mạch	multifactorial nature of cardiovascular risk	assess cardiovascular risk, cardiovascular risk factors
risk stratification	n	/rɪsk ˌstrætɪfɪˈkeɪʃən/	phân tầng rủi ro	enhanced risk stratification enables clinicians	improve risk stratification, risk stratification model
polypharmacy	n	/ˌpɒliˈfɑːməsi/	đa dược	polypharmacy affects millions of patients	manage polypharmacy, polypharmacy-related problems
continuous monitoring	n	/kənˈtɪnjuəs ˈmɒnɪtərɪŋ/	giám sát liên tục	AI-powered continuous monitoring systems	enable continuous monitoring, continuous monitoring devices
exacerbation	n	/ɪɡˌzæsəˈbeɪʃən/	đợt cấp, đợt bùng phát	predict disease exacerbations before they occur	prevent exacerbations, acute exacerbation
black box problem	n	/blæk bɒks ˈprɒbləm/	vấn đề hộp đen	the “black box” problem creates trust issues	address the black box problem, black box algorithms
medicolegal implications	n	/ˌmedɪkəʊˈliːɡəl ˌɪmplɪˈkeɪʃənz/	hàm ý y tế pháp lý	medicolegal implications of AI-assisted decisions	consider medicolegal implications, medicolegal concerns

Passage 3 – Essential Vocabulary

Từ vựng	Loại từ	Phiên âm	Nghĩa tiếng Việt	Ví dụ từ bài	Collocation
proliferation	n	/prəˌlɪfəˈreɪʃən/	sự gia tăng nhanh chóng	the proliferation of artificial intelligence	nuclear proliferation, weapons proliferation
precipitate	v	/prɪˈsɪpɪteɪt/	gây ra, thúc đẩy	has precipitated a paradigmatic shift	precipitate a crisis, precipitate change
engender	v	/ɪnˈdʒendə/	gây ra, tạo ra	engendering complex ethical dilemmas	engender trust, engender conflict
interdisciplinary scrutiny	n	/ˌɪntəˈdɪsɪplɪnəri ˈskruːtəni/	sự xem xét liên ngành	demand rigorous interdisciplinary scrutiny	require interdisciplinary scrutiny, interdisciplinary approach
pernicious	adj	/pəˈnɪʃəs/	có hại, nguy hiểm	one of the most pernicious challenges	pernicious effects, pernicious influence
perpetuate	v	/pəˈpetʃueɪt/	duy trì, làm tồn tại mãi	AI systems will perpetuate and amplify these inequities	perpetuate stereotypes, perpetuate inequality
underrepresented population	n	/ˌʌndəˌreprɪˈzentɪd ˌpɒpjuˈleɪʃən/	nhóm dân số thiếu đại diện	perform worse for underrepresented populations	serve underrepresented populations, underrepresented groups
proxy	n	/ˈprɒksi/	đại diện, chỉ số thay thế	use of healthcare costs as a proxy	serve as a proxy, proxy measure
epistemological	adj	/ɪˌpɪstɪməˈlɒdʒɪkəl/	nhận thức luận	epistemological challenges posed by AI	epistemological questions, epistemological framework
opacity	n	/əʊˈpæsəti/	sự mờ đục, không rõ ràng	through computational processes that are opaque	algorithmic opacity, opacity of decision-making
counterfactual explanation	n	/ˌkaʊntəˈfæktʃuəl ˌekspləˈneɪʃən/	giải thích phản thực tế	developing techniques such as counterfactual explanations	generate counterfactual explanations, counterfactual reasoning
granular	adj	/ˈɡrænjʊlə/	chi tiết, chi ly	require access to granular health information	granular data, granular level
longitudinal data	n	/ˌlɒndʒɪˈtjuːdɪnəl ˈdeɪtə/	dữ liệu theo thời gian	granular, longitudinal health information	collect longitudinal data, longitudinal study
re-identification attack	n	/ˌriːaɪˌdentɪfɪˈkeɪʃən əˈtæk/	tấn công xác định lại	re-identification attacks have demonstrated	prevent re-identification attacks, vulnerability to re-identification
exacerbate	v	/ɪɡˈzæsəbeɪt/	làm trầm trọng thêm	exacerbate or ameliorate existing health disparities	exacerbate inequality, exacerbate the problem
digital divide	n	/ˈdɪdʒɪtəl dɪˈvaɪd/	khoảng cách kỹ thuật số	the digital divide creates barriers	bridge the digital divide, narrow the digital divide
automation bias	n	/ˌɔːtəˈmeɪʃən ˈbaɪəs/	thiên lệch tự động hóa	automation bias has been documented	overcome automation bias, automation bias effect
fiduciary obligation	n	/fɪˈdjuːʃəri ˌɒblɪˈɡeɪʃən/	nghĩa vụ ủy thác	fiduciary obligations to shareholders	fulfill fiduciary obligations, fiduciary duty

---

Kết bài

Chủ đề “How is AI transforming personalized healthcare?” không chỉ là một trong những đề tài hot nhất trong IELTS Reading hiện nay mà còn phản ánh xu hướng phát triển quan trọng của y học hiện đại. Qua bộ đề thi mẫu hoàn chỉnh này, bạn đã được trải nghiệm đầy đủ cả ba cấp độ khó từ Easy đến Hard với 40 câu hỏi đa dạng, hoàn toàn giống với format thi thật.

Ba passages trong đề thi này đã cung cấp cho bạn góc nhìn toàn diện về AI trong y tế: từ những ứng dụng cơ bản và dễ hiểu ở Passage 1, đến các phân tích chuyên sâu về tối ưu hóa điều trị ở Passage 2, và cuối cùng là những thảo luận phức tạp về đạo đức và thách thức xã hội ở Passage 3. Độ khó tăng dần này không chỉ giúp bạn làm quen với cấu trúc đề thi mà còn rèn luyện khả năng đọc hiểu từ đơn giản đến phức tạp.

Phần đáp án chi tiết kèm theo giải thích cụ thể về vị trí thông tin, kỹ thuật paraphrase và cách xác định đáp án đúng sẽ giúp bạn tự đánh giá năng lực và hiểu rõ cách tiếp cận từng dạng câu hỏi. Đặc biệt, hệ thống từ vựng được phân loại theo từng passage với đầy đủ phiên âm, nghĩa, ví dụ và collocation sẽ giúp bạn không chỉ mở rộng vốn từ mà còn biết cách sử dụng chúng trong ngữ cảnh học thuật.

Hãy sử dụng đề thi này như một công cụ luyện tập thực chiến, thực hiện đúng thời gian quy định 60 phút để mô phỏng điều kiện thi thật. Sau khi hoàn thành, đối chiếu đáp án và đọc kỹ phần giải thích để hiểu rõ những điểm mạnh cần phát huy và điểm yếu cần cải thiện. Chúc bạn ôn tập hiệu quả và đạt được band điểm mục tiêu trong kỳ thi IELTS sắp tới!