The use of online obituaries as a tool for public health surveillance

项目摘要 本研究的目的是（1）建立不同年龄，性别和种族的讣告数据的代表性程度， 将这些信息与死亡证明记录进行比较，以了解开源数据的可靠性， 测量特性(2)建立一个模型，使用在线讣告数据来预测行政记录。 在卫生保健紧急情况下，必须监测全因死亡率，而不仅仅是特定原因的死亡 并计算多死亡人数的原因有几个：（1）官方统计的原因，具体 死亡人数可能会少算那些在死亡前没有检测出阳性的人;（2）医院和民事登记处可能不会 处理死亡证明需要几天甚至几周，这会造成数据滞后;（3）人 填写死亡证明书的人无法获得完整的医疗记录或以其他方式了解 （4）流行病和卫生紧急情况转移了人们的注意力和资源， 其它条件（例如，癌症患者已经看到延迟和推迟治疗）和气馁的人 在需要时不去医院（例如，中风），这可能间接导致增加， COVID-19以外疾病的死亡人数。从公开可用的文本挖掘中自动收集数据 在线讣告可以让我们快速预测死亡的年龄和性别分布， 成本效益的方式。目前，公开的数据集有两年的滞后期。从死亡记录的那一刻起 在这些人被释放之前被抓获，这种拖延妨碍了监测工作。提供资料说明 性别、年龄和种族是至关重要的，因为突发卫生事件可能直接或间接造成不成比例的 某些群体的死亡人数增加。在死亡率特别高（或低）的地方， 讣告数据，这种形式的监测可以告知政策反应的有效性。这项工作也可以 由于在线死亡记录更容易， 而且比管理数据更便宜。

Project Summary This study aims to (1) establish the degree of representativeness across age, sex, and race of obituary data by comparing that information with death certificate records to understand open-source data's reliability and measurement properties. (2) Build a model that uses online obituary data to predict administrative records. During health care emergencies, it is essential to monitor all-cause mortality and not just cause-specific deaths and to calculate the number of excess deaths for several reasons: (1) official statistics on cause-specific deaths might undercount people who did not test positive before dying; (2) hospitals and civil registries may not process death certificates for several days, or even weeks, which creates lags in the data; (3) the person completing the death certificate does not have access to the complete medical record or otherwise know about a positive test or symptoms; (4) pandemic and health emergencies divert attention and resources away from other conditions (e.g., cancer patients have seen delays and postponing treatment) and discouraged people from going to the hospital when needed (e.g., strokes), which may have indirectly caused an increase in fatalities from diseases other than COVID-19. Automated data collection from text mining of openly available online obituaries could allow us to derive quick predictions of age and sex distribution of death by location in a cost-effective way. Currently, publicly available datasets have a two-year lag. From the moment death records are captured to the time these are released, this delay hampers monitoring efforts. Providing information on sex, age, and race is critical because health emergencies might directly or indirectly cause a disproportionate increase in fatalities among certain groups. In places where mortality is exceptionally high (or low) based on obituary data, this form of monitoring can inform the policy response's effectiveness. This work can also be foundational for disease monitoring should future pandemics arise because online death records are easier and cheaper to access than administrative data.