Analysis of the Rationality of Medical Diagnostic Recommendations from the DeepSeek Model
Yucong Duan
Benefactor: Shiming Gong
International Standardization Committee of Networked DIKWP for Artificial Intelligence Evaluation(DIKWP-SC)
World Artificial Consciousness CIC(WAC)
World Conference on Artificial Consciousness(WCAC)
(Email: duanyucong@hotmail.com)
1Clinical logic errors
The diagnostic path given by DeepSeek has deviations in clinical reasoning. First, the patient has only coughed for about 1 week and some symptoms have been relieved, which is consistent with the typical course of acute bronchitis, but the model tends to complicate the process. It may over-interpret symptoms such as "night cough and night sweats" and regard them as clues to serious diseases such as tuberculosis. However, transient night sweats may also occur during acute inflammatory processes. Patients do not have long-term fever and weight loss, and their symptoms improve after antibiotic treatment. There is a lack of continuous progression characteristics of tuberculosis. At this time, common acute bronchitis should be considered first based on common sense, rather than abnormally giving priority to rare and severe diseases. This is a typical "diagnostic inversion". According to clinical guidelines, acute tracheobronchitis can be diagnosed as long as the patient has an acute onset, mainly coughs and sputum, and there is no other disease to explain the symptoms. DeepSeek did not fully follow this routine, complicating simple problems and violating the principle of common disease priority in clinical logic.
Secondly, DeepSeek's premise for judging the causal relationship of symptoms shows signs of error. For example, it may infer that the bacterial infection is still ongoing based on the color of sputum and recommend intensive treatment. However, the actual yellow-green sputum is not a reliable indicator of bacterial infection; the guidelines clearly state that "viral or bacterial infection cannot be judged based on the characteristics of sputum." The patient's sputum is thick but has improved under the treatment of cefaclor, indicating that the condition is improving and should not be inferred as a more serious infection. In addition, if the model ignores the fact that the patient has no high fever or obvious dyspnea and proposes a diagnosis of pneumonia, it is also a logical error. Pneumonia is usually manifested by higher fever, moist rales on lung auscultation, and imaging infiltration. This case does not have these signs, and there is no basis for rashly diagnosing pneumonia. Correct clinical thinking should attribute this case to the common diagnosis of acute bronchitis recovery period based on the severity of symptoms and the order of evolution, rather than jumping to rare or serious diseases.
2Improper medication recommendations
The DeepSeek model also has inappropriate recommendations for medication. The first is the use of antibiotics. The patient is a middle-aged male with no underlying disease, and his cough for a week is mostly self-limiting. The guidelines clearly do not recommend the routine use of antibiotics for acute simple bronchitis without pneumonia. Studies have shown that the efficacy of antibiotics for acute bronchitis is not obvious. If DeepSeek recommends continuing to use or replacing stronger antibiotics (such as upgrading to broad-spectrum antibiotics or extending the course of treatment), this is inconsistent with evidence-based medicine and will increase the risk of drug side effects and bacterial resistance. In this case, the patient's sputum has decreased under the action of cefaclor, indicating that no additional antibacterial drugs are needed. Excessive antibiotic treatment is not only unhelpful, but may also mask the condition or cause adverse reactions.
Secondly, the irrational use of hormone drugs may be another problem. If DeepSeek recommends the use of glucocorticoids (such as oral or inhaled hormones) to reduce cough airway inflammation, this is improper medication. There is currently no evidence to support that hormones can effectively treat cough caused by acute bronchitis. The use of hormones in patients with no history of asthma may not only increase side effects such as hyperglycemia and immunosuppression, but also does not help shorten the course of the disease. For this patient's night cough, the reasonable symptomatic treatment should be appropriate antitussive or expectorant drugs, rather than hormone drugs. If DeepSeek recommends inappropriate types or doses of drugs (such as using general cough suppressants for coughs with phlegm or using overdose drugs), it is improper medication advice and may endanger patient safety. Correct medication should be based on indications and evidence-based evidence: for example, when symptoms seriously affect rest, cough suppressants can be used to relieve night coughs as appropriate, and expectorants and fluid rehydration should be used during the day, and antibiotics or hormone interventions should not be abused.
3Diagnostic classification bias
DeepSeek's analysis may be biased in diagnostic classification, classifying the patient into the wrong disease category. This case actually tends to be in the recovery period of acute bronchitis, which is a self-limiting disease with a good prognosis; DeepSeek may mistakenly classify it as a chronic disease or serious condition. For example, if the model attributes the patient's transient cough aphonia and night cough to "chronic bronchitis" or "chronic obstructive pulmonary disease (COPD)", this is incorrect. The diagnosis of chronic bronchitis requires long-term recurrent coughing, at least 3 months a year, and more than 2 consecutive years. The patient has only been ill for one week and has no history of chronic cough, which obviously does not meet the classification criteria for chronic diseases. Similarly, if DeepSeek unilaterally considers "cough variant asthma" (only based on night cough), it also deviates from the mainstream judgment. Cough variant asthma generally has an allergic or asthma background, and is not accompanied by yellow-green sputum infection signs. The symptoms of this patient are more consistent with post-infection cough rather than asthma cough.
In addition, if the model interprets the patient's night sweats as tuberculosis or associates hoarseness with laryngeal tumors, it is also an over-diagnosis classification. Chronic infectious diseases such as pulmonary tuberculosis often cause coughs that linger for months and progressively worsen systemic symptoms such as weight loss and night sweats; in this case, the symptoms have been relieved in a short period of time and there is no sign of continued deterioration, so it should not be classified as a chronic disease such as tuberculosis. According to mainstream medical judgment, most patients with acute bronchitis have a good prognosis and symptoms will subside within a few weeks. If DeepSeek deviates from this point and misdiagnoses self-limiting diseases as critical or chronic diseases, it will undoubtedly cause unnecessary psychological burden and mislead patients. A more reasonable classification should be to recognize that this is the recovery stage of acute lower respiratory tract infection (bronchitis), similar to coughing after a common cold, rather than labeling the patient with a serious disease.
4Risk of over-medicalization
The recommendations of the DeepSeek model may also lead to over-medicalization, causing patients to receive unnecessary examinations and treatments. First of all, in terms of auxiliary examinations, if the model recommends a comprehensive imaging examination (such as a routine chest X-ray or CT) to "rule out" serious diseases, this is a manifestation of over-diagnosis and treatment. According to the primary diagnosis and treatment guidelines, routine chest imaging is not recommended for patients suspected of acute bronchitis unless there are warning signs such as hemoptysis, dyspnea, or signs of pulmonary consolidation. This patient did not have hemoptysis or obvious dyspnea. Blindly performing chest X-rays/CT will only increase radiation exposure and economic burden, but will hardly bring substantial benefits. If DeepSeek also recommends sputum culture, tuberculosis screening, or other complex laboratory tests, it is also inappropriate. For a case where symptoms have been relieved, large-scale screening for rare diseases is not only inefficient, but may also lead to unnecessary follow-up examinations due to false positive tests.
Secondly, in terms of treatment, the model's recommendations may have the problem of too fast escalation of measures. For example, the premature use of broad-spectrum antibiotics, intravenous infusion therapy, and even the combination of multiple drugs to "cover" all possible pathogens are typical over-medicalization tendencies. It is a consensus in the medical community that high-quality medical care should avoid overuse of antibiotics for acute bronchitis. The use of powerful drugs and measures without a target is not only unhelpful to recovery, but may also bring new drug side effects or iatrogenic problems. Similarly, if DeepSeek recommends that patients be referred to specialists immediately (such as asking patients to go to the respiratory department for bronchoscopy or to the ENT department for vocal cord examination), it is also unnecessary over-intervention. The guidelines point out that only when there is no improvement after conventional treatment or the cough persists for more than 8 weeks, further specialist evaluation is required to exclude other chronic causes. The patient in this case was far from reaching these levels, and rashly guiding him into a complex specialist diagnosis and treatment path would instead cause the patient to bear unnecessary examinations and psychological pressure during the recovery process. The reasonable approach should be to follow up and observe in a general outpatient clinic. If the symptoms continue to improve, no additional examination is required; only when the condition recurs or worsens, consider upgrading the diagnosis and treatment methods.
In summary, the recommendations given by the DeepSeek model have clinical decision-making bias. It failed to strictly follow evidence-based guidelines and unduly complicated a typical case of acute bronchitis recovery. In the context of public use, such recommendations may mislead patients in the following ways:
Exaggerating the severity of the disease: Associating a self-limited cough with a serious disease such as tuberculosis or tumors can cause unnecessary panic in patients. Patients may mistakenly believe that they have a serious illness and become overly anxious.
Improper medication: Incorrect medication recommendations (such as misuse of antibiotics or hormones) may lead patients to take unnecessary medications, which may lead to the risk of side effects and drug resistance. If patients follow such recommendations on their own, they may delay recovery due to inappropriate medications or even experience adverse drug reactions.
Excessive examination and treatment: Recommending a large number of examinations or specialist treatments will cause patients to waste time and money and bear unnecessary physical examination burdens. For example, inappropriate CT examinations increase radiation exposure, and early specialist visits may also cause patients to run around in unnecessary procedures.
Risk warning: For common conditions such as cough after a cold or acute bronchitis, avoid excessive stress and over-medicalization. Patients should observe changes in their condition under the guidance of a doctor, which can generally heal on its own or improve significantly within a few weeks. If you hear too serious a diagnosis or complex and expensive solutions from AI or the Internet, you may wish to be reserved and consult a professional doctor in time. It is a wise move to ensure safe recovery by following clinical guidelines for standardized diagnosis and treatment and avoiding unnecessary medications and examinations.
