Establishing an ad hoc COVID-19 mortality surveillance during the first epidemic wave in Belgium, 1 March to 21 June 2020

BackgroundCOVID-19-related mortality in Belgium has drawn attention for two reasons: its high level, and a good completeness in reporting of deaths. An ad hoc surveillance was established to register COVID-19 death numbers in hospitals, long-term care facilities (LTCF) and the community. Belgium adopted broad inclusion criteria for the COVID-19 death notifications, also including possible cases, resulting in a robust correlation between COVID-19 and all-cause mortality.AimTo document and assess the COVID-19 mortality surveillance in Belgium.MethodsWe described the content and data flows of the registration and we assessed the situation as of 21 June 2020, 103 days after the first death attributable to COVID-19 in Belgium. We calculated the participation rate, the notification delay, the percentage of error detected, and the results of additional investigations.ResultsThe participation rate was 100% for hospitals and 83% for nursing homes. Of all deaths, 85% were recorded within 2 calendar days: 11% within the same day, 41% after 1 day and 33% after 2 days, with a quicker notification in hospitals than in LTCF. Corrections of detected errors reduced the death toll by 5%.ConclusionBelgium implemented a rather complete surveillance of COVID-19 mortality, on account of a rapid investment of the hospitals and LTCF. LTCF could build on past experience of previous surveys and surveillance activities. The adoption of an extended definition of ‘COVID-19-related deaths’ in a context of limited testing capacity has provided timely information about the severity of the epidemic.     

In Vitro Characterization of In Situ Alloyed Ti6Al4V(ELI)-3 at.% Cu Obtained by Laser Powder Bed Fusion

The intensive cytotoxicity of pure copper is effectively kills bacteria, but it can compromise cellular behavior, so a rational balance must be found for Cu-loaded implants. In the present study, the individual and combined effect of surface composition and roughness on osteoblast cell behavior of in situ alloyed Ti6Al4V(ELI)-3 at.% Cu obtained by laser powder bed fusion was studied. Surface composition was studied using scanning electron microscopy, energy dispersive spectroscopy, and X-ray diffraction. Surface roughness measurements were carried out using confocal microscopy. In vitro osteoblast performance was evaluated by means of cell morphology observation of cell viability, proliferation, and mineralization. In vitro studies were performed at 1, 7, and 14 days of cell culture, except for cell mineralization at 28 days, on grounded and as-built (rough) samples with and without 3 at.% Cu. The addition of 3 at.% Cu did not show cell cytotoxicity but inhibited cell proliferation. Cell mineralization tends to be higher for samples with 3 at.% Cu content. Surface roughness inhibited cell proliferation too, but showed enhanced cell mineralization capacity and therefore, higher osteoblast performance, especially when as-built samples contained 3 at.% Cu. Cell proliferation was only observed on ground samples without Cu but showed the lowest cell mineralization.     

Beyond Cases and Deaths: The Benefits of Auxiliary Data Streams In Tracking the COVID-19 Pandemic: Can auxiliary indicators improve COVID-19 forecasting and hotspot prediction?

Short-term forecasts of traditional streams from public health reporting (such as cases, hospitalizations, and deaths) are a key input to public health decision-making during a pandemic. Since early 2020, our research group has worked with data partners to collect, curate, and make publicly available numerous real-time COVID-19 indicators, providing multiple views of pandemic activity in the United States. This paper studies the utility of five such indicators—derived from deidentified medical insurance claims, self-reported symptoms from online surveys, and COVID-related Google search activity—from a forecasting perspective. For each indicator, we ask whether its inclusion in an autoregressive (AR) model leads to improved predictive accuracy relative to the same model excluding it. Such an AR model, without external features, is already competitive with many top COVID-19 forecasting models in use today. Our analysis reveals that 1) inclusion of each of these five indicators improves on the overall predictive accuracy of the AR model; 2) predictive gains are in general most pronounced during times in which COVID cases are trending in “flat” or “down” directions; and 3) one indicator, based on Google searches, seems to be particularly helpful during “up” trends.

Tracking and forecasting indicators from public health reporting streams—such as confirmed cases and deaths in the COVID-19 pandemic—are crucial for understanding disease spread, correctly formulating public policy responses, and rationally planning future public health resource needs. A companion paper (1) describes our research group’s efforts, beginning in April 2020, in curating and maintaining a database of real-time indicators that track COVID-19 activity and other relevant phenomena. The signals (a term we use synonymously with “indicators”) in this database are accessible through the COVIDcast Application Programming Interface (API) (2), as well as associated R (3) and Python (4) packages, for convenient data fetching and processing. In the current paper, we quantify the utility provided by a core set of these indicators for two fundamental prediction tasks: probabilistic forecasting of COVID-19 case rates and prediction of future COVID-19 case hotspots (defined by the event that a relative increase in COVID-19 cases exceeds a certain threshold).At the outset, we should be clear that our intent in this paper is not to provide an authoritative take on cutting-edge COVID-19 forecasting methods. Similarly, some authors, e.g., ref. 5, have pointed out numerous mishaps of forecasting during the pandemic, and it is not our general intent to fix them here. Instead, we start with a basic and yet reasonably effective predictive model for future trends in COVID-19 cases and present a rigorous, quantitative assessment of the added value provided by auxiliary indicators that are derived from data sources that operate outside of traditional public health streams. In particular, we consider five indicators derived from deidentified medical insurance claims, self-reported symptoms from online surveys, and COVID-related Google searches.To assess this value in as direct terms as possible, we base our study around a very simple basic model: an autoregressive model, in which COVID cases in the near future are predicted using a linear combination of COVID cases in the near past. Forecasting carries a rich literature, offering a wide range of sophisticated techniques (see, e.g., ref. 6 for a review); however, we purposely avoid enhancements such as order selection, correction of outliers/anomalies in the data, and inclusion of regularization or nonlinearities. Similarly, we do not account for other factors that may well aid in forecasting, such as age-specific effects, holiday adjustments, and the effects of public health mandates. All that said, despite its simplicity, the basic autoregressive model that we consider in this paper exhibits competitive performance (see SI Appendix for details) with many of the top COVID-19 case forecasters submitted to the US COVID-19 Forecast Hub (7), which is the official source of forecasts used in public communications by the US Centers for Disease Control and Prevention (CDC). The strong performance of the autoregressive model here is in line with the fact that simple, robust models have also consistently been among the best-performing ones for COVID-19 death forecasting (8).In the companion paper (1), we analyze correlations between various indicators and COVID case rates. These correlations are natural summaries of the contemporaneous association between an indicator and COVID cases, but they fall short of delivering a satisfactory answer to the question that motivates the current article: Is the information contained in an indicator demonstrably useful for the prediction tasks we care about? Note that even lagged correlations cannot deliver a complete answer. Demonstrating utility for prediction is a much higher standard than simply asking about correlations; to be useful in forecast or hotspot models, an indicator must provide relevant information that is not otherwise contained in past values of the case rate series itself [cf. the pioneering work on Granger causality (9, 10), as well as the further references given below]. We assess this directly by inspecting the difference in predictive performance of simple autoregressive models trained with and without access to past values of a particular indicator.We find that each of the five indicators we consider—three based on COVID-related outpatient visits from medical insurance claims, one on self-reported symptoms from online surveys, and one on Google searches for anosmia or ageusia—provide an overall improvement in accuracy when incorporated into the autorgressive model. This is true both for COVID-19 case forecasting and for hotspot prediction. Further analysis reveals that the gains in accuracy depend on the pandemic’s dynamics at prediction time: The biggest gains in accuracy appear during times in which cases are “flat” or trending “down”; but the indicator based on Google searches offers a notable improvement when cases are trending “up.”Careful handling of data revisions plays a key role in our analysis. Signals computed from surveillance streams are often subject to latency and/or revision. For example, a signal based on aggregated medical insurance claims may be available after just a few days, but it can then be substantially revised over the next several weeks as additional claims are submitted and/or processed late. Correlations between such a signal and case rates calculated “after the fact” (i.e., computed retrospectively, using the finalized values of this signal) will not deliver an honest answer to the question of whether this signal would have been useful in real time. Instead, we build predictive models using only the data that would have been available as of the prediction date and compare the ensuing predictions in terms of accuracy. The necessity of real-time data for honest forecast evaluations has been recognized in econometrics for a long time (11–21), but it is often overlooked in epidemic forecasting despite its critical importance (22).Finally, it is worth noting that examining the importance of additional features for prediction is a core question in inferential statistics and econometrics, with work dating back to at least ref. 9. Still today, drawing rigorous inference based on predictions, without (or with lean) assumptions, is an active field of research from both applied and theoretical angles (23–32). Our take in the current work is in line with much of this literature; however, to avoid making any explicit assumptions, we do not attempt to make formal significance statements and, instead, broadly examine the stability of our conclusions with respect to numerous modes of analysis.     

The Consumption of Food-Based Iodine in the Immediate Pre-Pregnancy Period in Madrid Is Insufficient. San Carlos and Pregnancy Cohort Study

A pre-gestational thyroid reserve of iodine is crucial to guarantee the increased demand for thyroid hormone production of early pregnancy. An iodine intake ≥150 µg/day is currently recommended. The objective of this study was to assess average pre-gestational food-based iodine consumption in pregnant women at their first prenatal visit (<12 gestational weeks), and its association with adverse materno-fetal events (history of miscarriages, early fetal losses, Gestational Diabetes, prematurity, caesarean sections, and new-borns large/small for gestational age). Between 2015–2017, 2523 normoglycemic women out of 3026 eligible had data in the modified Diabetes Nutrition and Complication Trial (DNCT) questionnaire permitting assessment of pre-gestational food-based iodine consumption, and were included in this study. Daily food-based iodine intake was 123 ± 48 µg, with 1922 (76.1%) not reaching 150 µg/day. Attaining this amount was associated with consuming 8 weekly servings of vegetables (3.84; 3.16–4.65), 1 of shellfish (8.72; 6.96–10.93) and/or 2 daily dairy products (6.43; 5.27–7.86). Women who reached a pre-gestational intake ≥150 µg had lower rates of hypothyroxinemia (104 (17.3%)/384 (21.4%); p = 0.026), a lower miscarriage rate, and a decrease in the composite of materno-fetal adverse events (0.81; 0.67–0.98). Reaching the recommended iodine pre-pregnancy intake with foods could benefit the progression of pregnancy.     

Aberrant TIMP-1 production in tumor-associated fibroblasts drives the selective benefits of nintedanib in lung adenocarcinoma

The fibrotic tumor microenvironment is a pivotal therapeutic target. Nintedanib, a clinically approved multikinase antifibrotic inhibitor, is effective against lung adenocarcinoma (ADC) but not squamous cell carcinoma (SCC). Previous studies have implicated the secretome of tumor-associated fibroblasts (TAFs) in the selective effects of nintedanib in ADC, but the driving factor(s) remained unidentified. Here we examined the role of tissue inhibitor of metalloproteinase-1 (TIMP-1), a tumor-promoting cytokine overproduced in ADC-TAFs. To this aim, we combined genetic approaches with in vitro and in vivo preclinical models based on patient-derived TAFs. Nintedanib reduced TIMP-1 production more efficiently in ADC-TAFs than SCC-TAFs through a SMAD3-dependent mechanism. Cell culture experiments indicated that silencing TIMP1 in ADC-TAFs abolished the therapeutic effects of nintedanib on cancer cell growth and invasion, which were otherwise enhanced by the TAF secretome. Consistently, co-injecting ADC cells with TIMP1-knockdown ADC-TAFs into immunocompromised mice elicited a less effective reduction of tumor growth and invasion under nintedanib treatment compared to tumors bearing unmodified fibroblasts. Our results unveil a key mechanism underlying the selective mode of action of nintedanib in ADC based on the excessive production of TIMP-1 in ADC-TAFs. We further pinpoint reduced SMAD3 expression and consequent limited TIMP-1 production in SCC-TAFs as key for the resistance of SCC to nintedanib. These observations strongly support the emerging role of TIMP-1 as a critical regulator of therapy response in solid tumors.     

Yoga for symptom management in oncology: A review of the evidence base and future directions for research

Because yoga is increasingly recognized as a complementary approach to cancer symptom management, patients/survivors and providers need to understand its potential benefits and limitations both during and after treatment. The authors reviewed randomized controlled trials (RCTs) of yoga conducted at these points in the cancer continuum (N = 29; n = 13 during treatment, n = 12 post-treatment, and n = 4 with mixed samples). Findings both during and after treatment demonstrated the efficacy of yoga to improve overall quality of life (QOL), with improvement in subdomains of QOL varying across studies. Fatigue was the most commonly measured outcome, and most RCTs conducted during or after cancer treatment reported improvements in fatigue. Results also suggested that yoga can improve stress/distress during treatment and post-treatment disturbances in sleep and cognition. Several RCTs provided evidence that yoga may improve biomarkers of stress, inflammation, and immune function. Outcomes with limited or mixed findings (eg, anxiety, depression, pain, cancer-specific symptoms, such as lymphedema) and positive psychological outcomes (such as benefit-finding and life satisfaction) warrant further study. Important future directions for yoga research in oncology include: enrolling participants with cancer types other than breast, standardizing self-report assessments, increasing the use of active control groups and objective measures, and addressing the heterogeneity of yoga interventions, which vary in type, key components (movement, meditation, breathing), dose, and delivery mode.     

Altered function of the glutamate–aspartate transporter GLAST,a potential therapeutic target in glioblastoma

In glioma patients, high levels of glutamate can cause brain edema and seizures. GLAST, a glutamate–aspartate transporter expressed by astrocytes with a role in glutamate uptake, is highly expressed on the plasma membrane of glioblastoma (GBM) cells, and its expression significantly correlates with shortened patient survival. Here, it was demonstrated that inhibition of GLAST expression limited the progression and invasion of GBM xenografts. Magnetic resonance spectroscopy was used to measure glutamate in GLAST-expressing gliomas showing that these tumors exhibit increased glutamate concentration compared to GLAST-depleted glioma. Despite their GLAST expression, GBM stem-like cells (GSCs) released rather than taking up glutamate due to their lack of Na+/K+-ATPase. Overexpression of Na+/K+-ATPase in these cells restored glutamate uptake and induced apoptosis. The therapeutic relevance of targeting GLAST in gliomas was assessed using the inhibitor UCPH-101. In glioma-bearing mice, a single intratumoral injection of UCPH-101 significantly increased survival by decreasing GLAST expression and inducing apoptosis. Thus, GLAST has a novel role in GBM that appears to have crucial relevance in glutamate trafficking and may thus be a new therapeutic target.     

Somatic mutations and promotor methylation of the ryanodine receptor 2 is a common event in the pathogenesis of head and neck cancer

Genomic sequencing projects unraveled the mutational landscape of head and neck squamous cell carcinoma (HNSCC) and provided a comprehensive catalog of somatic mutations. However, the limited number of significant cancer-related genes obtained so far only partially explains the biological complexity of HNSCC and hampers the development of novel diagnostic biomarkers and therapeutic targets. We pursued a multiscale omics approach based on whole-exome sequencing, global DNA methylation and gene expression profiling data derived from tumor samples of the HIPO-HNC cohort (n = 87), and confirmed new findings with datasets from The Cancer Genome Atlas (TCGA). Promoter methylation was confirmed by MassARRAY analysis and protein expression was assessed by immunohistochemistry and immunofluorescence staining. We discovered a set of cancer-related genes with frequent somatic mutations and high frequency of promoter methylation. This included the ryanodine receptor 2 (RYR2), which showed variable promoter methylation and expression in both tumor samples and cell lines. Immunohistochemical staining of tissue sections unraveled a gradual loss of RYR2 expression from normal mucosa via dysplastic lesion to invasive cancer and indicated that reduced RYR2 expression in adjacent tissue and precancerous lesions might serve as risk factor for unfavorable prognosis and upcoming malignant conversion. In summary, our data indicate that impaired RYR2 function by either somatic mutation or epigenetic silencing is a common event in HNSCC pathogenesis. Detection of RYR2 expression and/or promoter methylation might enable risk assessment for malignant conversion of dysplastic lesions.     

Statin use and pancreatic cancer risk in two prospective cohort studies

Background

Statins, 3-hydroxy-3-methylglutaryl-coenzyme A (HMG-CoA) reductase inhibitors, are common lipid-lowering agents and may reduce the risk of several cancer types including pancreatic cancer. However, the association between statin use and pancreatic cancer risk has not been fully evaluated in prospective studies.

Methods

We studied the association between statin use and incident pancreatic cancer in 113,059 participants from the prospective Nurses’ Health Study and Health Professionals Follow-up Study. Statin use was self-reported via study questionnaires and updated biennially. Hazard ratios (HRs) and 95% confidence intervals (CIs) for incidence of pancreatic cancer were estimated using multivariable Cox proportional hazards models with adjustment for potential confounders.

Results

In total, 583 participants developed incident pancreatic cancer during 1.4 million person-years of follow-up. No difference was identified in pancreatic cancer risk for regular versus non-regular statin users (multivariable-adjusted HR 0.98; 95% CI 0.82–1.16). There was no significant heterogeneity in the association of statin use with pancreatic cancer risk between the cohorts. Similarly, longer duration of regular statin use was not associated with decreased risk of pancreatic cancer (P_trend = 0.65). The results remained similar when we examined statin use status at baseline or accounting for 4-year latency period. We observed no statistically significant effect modification for the association of statin use with pancreatic cancer risk by body mass index, smoking status, or diabetes mellitus status (all P_interaction > 0.21).

Conclusions

Regular statin use was not associated with pancreatic cancer risk in two large prospective cohort studies in the U.S.