Using Machine Learning to support health system planning during the Covid-19 pandemic: a case study using data from São José dos Campos (Brazil)

Many efforts were made by the scientific community during the Covid-19 pandemic to understand the disease and better manage health systems' resources. Believing that city and population characteristics influence how the disease spreads and develops, we used Machine Learning techniques to provide insights to support decision-making in the city of São José dos Campos (SP), Brazil. Using a database with information from people who undergo the Covid-19 test in this city, we generate and evaluate predictive models related to severity, need for hospitalization and period of hospitalization. Additionally, we used the SHAP value for models' interpretation of the most decisive attributes influencing the predictions. We can conclude that patient age linked to symptoms such as saturation and respiratory distress and comorbidities such as cardiovascular disease and diabetes are the most important factors to consider when one wants to predict severity and need for hospitalization in this city. We also stress the need of a greater attention to the proper collection of this information from citizens who undergo the Covid-19 diagnosis test.

Delayed Colorectal Cancer Diagnosis during the COVID-19 Pandemic in Alberta: A Framework for Analyzing Barriers to Diagnosis and Generating Evidence to Support Health System Changes Aimed at Reducing Time to Diagnosis

The frequency of colorectal cancer (CRC) diagnosis has decreased due to the COVID-19 pandemic. Health system planning is needed to address the backlog of undiagnosed patients. We developed a framework for analyzing barriers to diagnosis and estimating patient volumes under different system relaunch scenarios. This retrospective study included CRC cases from the Alberta Cancer Registry for the pre-pandemic (1 January 2016–4 March 2020) and intra-pandemic (5 March 2020–1 July 2020) periods. The data on all the diagnostic milestones in the year prior to a CRC diagnosis were obtained from administrative health data. The CRC diagnostic pathways were identified, and diagnostic intervals were measured. CRC diagnoses made during hospitalization were used as a proxy for severe disease at presentation. A modified Poisson regression analysis was used to estimate the adjusted relative risk (adjRR) and a 95% confidence interval (CI) for the effect of the pandemic on the risk of hospital-based diagnoses. During the study period, 8254 Albertans were diagnosed with CRC. During the pandemic, diagnosis through asymptomatic screening decreased by 6·5%. The adjRR for hospital-based diagnoses intra-COVID-19 vs. pre-COVID-19 was 1.24 (95% CI: 1.03, 1.49). Colonoscopies were identified as the main bottleneck for CRC diagnoses. To clear the backlog before progression is expected, high-risk subgroups should be targeted to double the colonoscopy yield for CRC diagnosis, along with the need for a 140% increase in monthly colonoscopy volumes for a period of 3 months. Given the substantial health system changes required, it is unlikely that a surge in CRC cases will be diagnosed over the coming months. Administrators in Alberta are using these findings to reduce wait times for CRC diagnoses and monitor progression.

Cancer care in the time of COVID-19: A conceptual framework to improve cancer outcomes during a pandemic.

e18628 Background: At the start of the COVID-19 pandemic, a plan for cancer management during a pandemic did not exist. It soon became clear that without proper planning, cancer outcomes would worsen. Cancer patients are at increased risk of COVID-19 infection, morbidity, and mortality. Health sectors internationally reduced or paused non-urgent cancer care to protect cancer patients from COVID-19. However, disproportionate delays in screening, diagnosis, and treatment can unduly impact cancer outcomes, and backlogs can further burden a strained health system. Tailored approaches to cancer management are required which balance health resource availability along with the risks of exposure and benefits of treatment. Australia’s relatively low COVID-19 case numbers afforded Cancer Australia an opportunity to proactively plan for optimal cancer management during this, and future, pandemics. Methods: Cancer Australia’s Cancer care in the time of COVID-19: A conceptual framework for the management of cancer during a pandemic (the framework) maps evidence-based cancer care considerations in relation to a health system’s capacity across acute and recovery pandemic phases, in relation to steps of the cancer care pathway. The framework promotes infection control and resource prioritisation in the context of innovative care models, triaging approaches and individualised treatment plans, underpinned by effective communication and shared decision-making. Results: The framework supports health system planning and risk-stratified approaches to guide decision-making and improve cancer outcomes. Many aspects of cancer care are recommended to continue (to varying degrees) in most pandemic phases, with modifications or pauses in some aspects of care as the pandemic curve approaches or exceeds health system capacity. Principles of the framework were employed during the second wave of COVID-19 in the Australian state of Victoria, with continuation of cancer screening programs, diagnostic investigations, and treatments wherever it was safe to do so. This resulted in reductions in cancer services and treatment being relatively smaller than in the first wave. Conclusions: Cancer management in a pandemic is not a one-size-fits-all. Countries and jurisdictions need to tailor cancer care according to the risk of the health system becoming overwhelmed. The framework guides optimal cancer care to improve outcomes for people with cancer, while minimising COVID-19 infection. As further evidence becomes available from this pandemic or in future pandemics, this framework can be refined to inform ongoing and future pandemic health system planning.

Using Standardized Electronic Pathology and Surgery Data to Inform Clinical Quality Improvement and Health System Planning

Background: Practice variation in diagnosis and treatment exists between clinicians and jurisdictions across Canada. This variation can impact the quality of care that patients receive and patient outcomes. Knowledge of the scale and type of variation is the first step to developing action plans to improve consistency and enhance patient care. Aim: We aimed to establish a method by which to examine the magnitude of practice variation between clinicians and interjurisdictionally within the cancer system. We leveraged and derived evidence from discrete pathology data collected by five Canadian jurisdictions at the point of care to identify areas to improve quality of cancer care services and to direct patient care. Methods: Fifty pathologists, surgeons, and medical oncologists from 10 jurisdictions conferred to leverage literature and data standards (developed by the College of American Pathologists (CAP)) to create 48 descriptive and outcome indicators related to five cancers: breast, lung, colorectal, endometrial, and prostate cancer. Five jurisdictions collected and used data to generate the indicators. This baseline data were reviewed by 65 clinicians. Results: Interjurisdictional comparative baseline data analyses on 48 indicators showed clinical validity and relevance for use to direct downstream patient care. Data characterizing cancer type, stage, and grade distribution were consistently reported across geography and aligned with the evidence noted in the literature. The data also noted practice and performance variation across multiple cancer sites. For example, although the recommended guideline is to examine at least 12 lymph nodes in 90% of colorectal cancer patients, only one province met this target. Another example is Lynch syndrome testing, which may be important for patients with a diagnosis of colorectal or endometrial cancer depending on the age at diagnosis and family history. The data showed that 0%–70% of patients diagnosed with colorectal cancer prior to age 70 received testing for Lynch syndrome, and only 10%–40% of endometrial cancer cases were tested for markers of Lynch syndrome across the country. The value of these indicators is enormous to inform potential training opportunities and set standards of care at the local or broader clinical governance level so that consistent, high-quality care is delivered in accordance with evidence-based guidelines. Conclusion: Practice variation exists between clinicians and jurisdictions, and comparative pathology data can be used to create a cancer learning system. Four jurisdictions are now embarking on leveraging indicator data analysis to generate physician-level feedback reports and convening communities of practice with the goal of facilitating peer-to-peer conversations, and establishing benchmarks and targets to improve the quality of care, refine or develop clinical guidelines, and inform health system planning in Canada. These lessons can be applied in other cancer systems.