Objective: We examined sociobehavioural factors associated with SARS-CoV-2 infection and estimated COVID-19 vaccine effectiveness against symptomatic SARS-CoV-2 infection in the Philippines. Such studies are limited in low- and middle-income countries, especially in Asia and the Pacific. Methods: A case-control study was conducted in two hospitals in Manila, Philippines, from March 2022 to June 2023. Sociobehavioural factors and vaccination history were collected. PCR-positive individuals were cases, while PCR-negative individuals were controls. Adjusted odds ratios (aORs) were calculated to examine associations between sociobehavioural factors/vaccination and medically attended SARS-CoV-2 infection. Results: The analysis included 2489 individuals (574 positive cases, 23.1%; 1915 controls, 76.9%; median age [interquartile range]: 35 [27–51] years). Although education and household income were not associated with infection, being a health-care worker was (aOR: 1.45; 95% confidence interval [CI]: 1.03–2.06). The odds of infection were higher among individuals who attended gatherings of five or more people compared to those who attended smaller gatherings (aOR: 2.58; 95% CI: 1.14–5.83). Absolute vaccine effectiveness for vaccination status was not estimated due to a high risk of bias, for example, unascertained prior infection. Moderate relative vaccine effectiveness for the first booster (32%; 95% CI: -120–79) and the second booster (48%; 95% CI: -23–78) were observed (both with wide CI), albeit with a waning trend after half a year. Discussion: The higher odds of infection among health-care workers emphasize the importance of infection prevention and control measures. Moderate relative vaccine effectiveness with a waning trend reiterates the need for more efficacious vaccines against symptomatic infection caused by circulating variants and with longer duration of protection.

Objectives: Developing large language models (LLMs) in biomedicine requires access to high-quality training and alignment tuning datasets. However, publicly available Korean medical preference datasets are scarce, hindering the advancement of Korean medical LLMs. This study constructs and evaluates the efficacy of the Korean Medical Preference Dataset (KoMeP), an alignment tuning dataset constructed with an automated pipeline, minimizing the high costs of human annotation. Methods: KoMeP was generated using the DAHL score, an automated hallucination evaluation metric. Five LLMs (Dolly-v2-3B, MPT-7B, GPT-4o, Qwen-2-7B, Llama-3-8B) produced responses to 8,573 biomedical examination questions, from which 5,551 preference pairs were extracted. Each pair consisted of a “chosen” response and a “rejected” response, as determined by their DAHL scores. The dataset was evaluated when trained through two different alignment tuning methods, direct preference optimization (DPO) and odds ratio preference optimization (ORPO) respectively across five different models. The KorMedMCQA benchmark was employed to assess the effectiveness of alignment tuning. Results: Models trained with DPO consistently improved KorMedMCQA performance; notably, Llama-3.1-8B showed a 43.96% increase. In contrast, ORPO training produced inconsistent results. Additionally, English-to-Korean transfer learning proved effective, particularly for English-centric models like Gemma-2, whereas Korean-to-English transfer learning achieved limited success. Instruction tuning with KoMeP yielded mixed outcomes, which suggests challenges in dataset formatting. Conclusions KoMeP is the first publicly available Korean medical preference dataset and significantly improves alignment tuning performance in LLMs. The DPO method outperforms ORPO in alignment tuning. Future work should focus on expanding KoMeP, developing a Korean-native dataset, and refining alignment tuning methods to produce safer and more reliable Korean medical LLMs.

Objectives: Developing large language models (LLMs) in biomedicine requires access to high-quality training and alignment tuning datasets. However, publicly available Korean medical preference datasets are scarce, hindering the advancement of Korean medical LLMs. This study constructs and evaluates the efficacy of the Korean Medical Preference Dataset (KoMeP), an alignment tuning dataset constructed with an automated pipeline, minimizing the high costs of human annotation. Methods: KoMeP was generated using the DAHL score, an automated hallucination evaluation metric. Five LLMs (Dolly-v2-3B, MPT-7B, GPT-4o, Qwen-2-7B, Llama-3-8B) produced responses to 8,573 biomedical examination questions, from which 5,551 preference pairs were extracted. Each pair consisted of a “chosen” response and a “rejected” response, as determined by their DAHL scores. The dataset was evaluated when trained through two different alignment tuning methods, direct preference optimization (DPO) and odds ratio preference optimization (ORPO) respectively across five different models. The KorMedMCQA benchmark was employed to assess the effectiveness of alignment tuning. Results: Models trained with DPO consistently improved KorMedMCQA performance; notably, Llama-3.1-8B showed a 43.96% increase. In contrast, ORPO training produced inconsistent results. Additionally, English-to-Korean transfer learning proved effective, particularly for English-centric models like Gemma-2, whereas Korean-to-English transfer learning achieved limited success. Instruction tuning with KoMeP yielded mixed outcomes, which suggests challenges in dataset formatting. Conclusions KoMeP is the first publicly available Korean medical preference dataset and significantly improves alignment tuning performance in LLMs. The DPO method outperforms ORPO in alignment tuning. Future work should focus on expanding KoMeP, developing a Korean-native dataset, and refining alignment tuning methods to produce safer and more reliable Korean medical LLMs.

Objectives: Developing large language models (LLMs) in biomedicine requires access to high-quality training and alignment tuning datasets. However, publicly available Korean medical preference datasets are scarce, hindering the advancement of Korean medical LLMs. This study constructs and evaluates the efficacy of the Korean Medical Preference Dataset (KoMeP), an alignment tuning dataset constructed with an automated pipeline, minimizing the high costs of human annotation. Methods: KoMeP was generated using the DAHL score, an automated hallucination evaluation metric. Five LLMs (Dolly-v2-3B, MPT-7B, GPT-4o, Qwen-2-7B, Llama-3-8B) produced responses to 8,573 biomedical examination questions, from which 5,551 preference pairs were extracted. Each pair consisted of a “chosen” response and a “rejected” response, as determined by their DAHL scores. The dataset was evaluated when trained through two different alignment tuning methods, direct preference optimization (DPO) and odds ratio preference optimization (ORPO) respectively across five different models. The KorMedMCQA benchmark was employed to assess the effectiveness of alignment tuning. Results: Models trained with DPO consistently improved KorMedMCQA performance; notably, Llama-3.1-8B showed a 43.96% increase. In contrast, ORPO training produced inconsistent results. Additionally, English-to-Korean transfer learning proved effective, particularly for English-centric models like Gemma-2, whereas Korean-to-English transfer learning achieved limited success. Instruction tuning with KoMeP yielded mixed outcomes, which suggests challenges in dataset formatting. Conclusions KoMeP is the first publicly available Korean medical preference dataset and significantly improves alignment tuning performance in LLMs. The DPO method outperforms ORPO in alignment tuning. Future work should focus on expanding KoMeP, developing a Korean-native dataset, and refining alignment tuning methods to produce safer and more reliable Korean medical LLMs.

roblem: Once COVID-19 vaccines were rolled out, there was a need to monitor real-world vaccine effectiveness to accumulate evidence to inform policy and risk communication. This was especially true in Japan and the Philippines, given historical issues that affected vaccine confidence. Context: Neither country had public health surveillance that could be enhanced to evaluate vaccine effectiveness or readily available national vaccination databases. Action: Study groups were established in multiple health-care facilities in each country to assess vaccine effectiveness against both symptomatic infection and severe disease. Outcome: In Japan, multiple study reports were published in Japanese on the website of the National Institute of Infectious Diseases and presented at the national government’s advisory board. Nationwide media coverage facilitated transparency and increased the confidence of the government and the public in the vaccination programme. In the Philippines, the launch of the study was delayed so as to align the research plan with the interests of various stakeholders and to obtain institutional review board approval. Ultimately, the studies were successfully initiated and completed. Discussion: There were four main challenges in conducting our studies: finding health-care facilities for data collection; obtaining exposure (vaccination) data; identifying epidemiological biases and confounders; and informing policy and risk communication in a timely manner. Preparedness during inter-emergency/epidemic/pandemic periods to rapidly evaluate relevant interventions such as vaccination is critical and should include the following considerations: (1) the establishment and maintenance of prospective data collection platforms, ideally under public health surveillance (if not, clinical research networks or linked databases); (2) uniform and practical protocols considering biases and confounders; and (3) communication with stakeholders including institutional review boards.

Background and Objectives: Blood pressure (BP) measurement using an auscultatory sphygmomanometer is recommended for diagnosing hypertension in children. As mercury sphygmomanometers (MSs) are banned owing to environmental concerns, it is crucial to determine the accuracy of mercury-free sphygmomanometers to replace them. We analyzed the accuracy of these devices to guide the National Survey selection. Methods: BP was measured thrice each with MS, auscultatory device (AD), and oscillometric device (OD) in 104 participants aged 10–18 using the National Survey data. The difference in BP was defined as the difference between MS and other devices. The BP differences, correlations, and influencing factors were analyzed. The frequencies of hypertension were also compared. Results: Systolic BP (SBP) and diastolic BP (DBP) differences between MS and AD were 0.88±3.36 mmHg and 0.63±3.95 mmHg, and those between MS and OD were 0.43±5.83 mmHg and 4.57±6.89 mmHg, respectively. The absolute error of <10 mmHg for DBP between MS and OD was 76%. The concordance correlation coefficient between MS and AD was 0.94 for SBP and 0.90 for DBP, and 0.81 and 0.67, respectively for MS and OD. Arm circumference negatively correlated with BP differences except for SBP between the MS and OD. The frequency of hypertension was not different between MS and AD but was underestimated by OD. Conclusions AD correlated well with MS, while OD did not, especially for DBP. The superiority of AD over OD suggests AD as a possible alternative for MS in the National Survey.