Surveillance bias in the assessment of the size of COVID-19 epidemic waves: a case study.

OBJECTIVES: To estimate the size of COVID-19 waves using four indicators across three pandemic periods and assess potential surveillance bias. STUDY DESIGN: Case study using data from one region of Switzerland. METHODS: We compared cases, hospitalizations, deaths, and seroprevalence during three periods including the first three pandemic waves (period 1: Feb-Oct 2020; period 2: Oct 2020-Feb 2021; period 3: Feb-Aug 2021). Data were retrieved from the Federal Office of Public Health or estimated from population-based studies. To assess potential surveillance bias, indicators were compared to a reference indicator, i.e. seroprevalence during periods 1 and 2 and hospitalizations during the period 3. Timeliness of indicators (the duration from data generation to the availability of the information to decision-makers) was also evaluated. RESULTS: Using seroprevalence (our reference indicator for period 1 and 2), the 2nd wave size was slightly larger (by a ratio of 1.4) than the 1st wave. Compared to seroprevalence, cases largely overestimated the 2nd wave size (2nd vs 1st wave ratio: 6.5), while hospitalizations (ratio: 2.2) and deaths (ratio: 2.9) were more suitable to compare the size of these waves. Using hospitalizations as a reference, the 3rd wave size was slightly smaller (by a ratio of 0.7) than the 2nd wave. Cases or deaths slightly underestimated the 3rd wave size (3rd vs 2nd wave ratio for cases: 0.5; for deaths: 0.4). The seroprevalence was not useful to compare the size of these waves due to high vaccination rates. Across all waves, timeliness for cases and hospitalizations was better than for deaths or seroprevalence. CONCLUSIONS: The usefulness of indicators for assessing the size of pandemic waves depends on the type of indicator and the period of the pandemic.

Home Visiting and Child Welfare Involvement: A Matched Comparison Group Study.

The present study is one of the largest quasi-experimental studies to date on the effects of home visiting on documented child maltreatment during a child's first two years of life. In this matched comparison group study, we compare 8796 families that participated in a home visiting program (HV families) to 8796 similar non-participating families (non-HV families) selected from birth records using Coarsened Exact Matching. Using sequential logistic regression, we identify that HV families have significantly higher odds of experiencing a child maltreatment investigation by their child's second birthday compared to non-HV families; however, among those that were investigated, HV families have significantly lower odds of having their first investigation substantiated for maltreatment. Overall, HV families do not differ significantly from non-HV families in the odds of experiencing a substantiated investigation over 2 years. We share implications for considering surveillance bias, and we highlight the importance of including both substantiated and unsubstantiated investigations when studying the effects of home visiting on documented child maltreatment.

Letter to the Editor: Pathogens detected from patients with acute respiratory infections negative for SARS-CoV-2, Saitama, Japan, 2020

@#Utilizing data presented in the article by Miyashita et al., we illustrate the importance of testing data when assessing surveillance data. Accounting for the number of tests (denominator) and positivity (proportion of tests positive for a specific pathogen(s)) improves data interpretation in ways not possible from numerator case data alone.

Cancer Risk in Patients with Gaucher Disease Using Real-World Data.

The association between GD and cancer has been uncertain due to ascertainment bias in previously published studies. We analyzed cancer incidence using the Maccabi Healthcare Service (MHS) electronic health records among 264 patients with GD compared to 3440 matched controls. We ascertained cancers diagnosed before and after the index date (i.e., the first documentation of GD in cases and the corresponding date for controls). Before the index date, cancers were diagnosed in 18 individuals, with 11 (4.2%) in the GD group and 7 (0.2%) in the control group. After the index date, cancers were diagnosed in 57 individuals, with 20 (7.9%) in the GD group and 37 (1.1%) in the control group, with a median follow-up of almost 13 years in both groups. The most common cancers diagnosed in GD were non-melanoma skin cancer (NMSC) and hematological malignancies, with a clustering of diagnoses around the time of GD diagnosis. The incidence of cancers (excluding MNSC) was 4.1 (95% CI 2.2-7.1) and 0.7 (95% CI 0.4-0.9) per 1000 patient-years in the GD and control groups, respectively, with an incidence rate ratio of 6.37 (95% CI 3-12.7). Patients with GD underwent more cancer screening tests than their counterparts in the control group. While our study revealed an increased occurrence of cancers in patients with GD, this finding might be partly attributed to the more rigorous surveillance procedures employed in this patient population.

Association Between COVID-19 Infection and Pulmonary Fibrosis: A Nested Case-Control Study.

BACKGROUND: Pulmonary fibrosis is associated with significant morbidity. Data are scarce on the link between coronavirus disease (COVID-19) and pulmonary fibrosis. We aimed to assess the association between COVID-19 with pulmonary fibrosis. METHODS: We conducted a nested case-control study in a cohort of 2,894,801 adults without a diagnosis of pulmonary fibrosis. The underlying cohort consisted of members of the largest healthcare provider in Israel aged 18 years or older as of May 1, 2020. Subjects were followed up from cohort entry until June 30, 2022, for the occurrence of pulmonary fibrosis. Ten randomly selected controls were matched to each case of pulmonary fibrosis on age, sex, and calendar time. To account for surveillance bias a lag time of 60 days was used for ascertainment of prior COVID-19 and COVID-19 severity. RESULTS: During follow-up 1284 patients were newly diagnosed with pulmonary fibrosis and matched with 12,840 controls. Multivariable conditional logistic-regression models showed that the odds ratio for pulmonary fibrosis was 1.80 (95% confidence interval, 1.47-2.19) in patients with COVID-19 compared with no COVID-19. The multivariable odds ratio for pulmonary fibrosis was 1.33 (1.06-1.68), 2.98 (1.16-7.65), and 9.30 (5.77-14.98) for mild, moderate, and severe COVID-19, respectively, compared with no COVID-19. The magnitude of the association was attenuated but remained statistically significant for severe disease when the lag time was extended to 180 days (1.08 [0.78-1.49], 2.37 [0.75-7.46], and 5.34 [2.75-10.36] for mild, moderate, and severe COVID-19, respectively). CONCLUSIONS: COVID-19 appears to be associated with an increased risk of pulmonary fibrosis and the magnitude of the association increases with COVID-19 severity.

Association between SARS-CoV-2 infection and new-onset atrial fibrillation.

BACKGROUND: Atrial fibrillation (AF) is associated with substantial morbidity and mortality. New-onset AF (NOAF) has been related recently to SARS-CoV-2 infection; however, the evidence supporting this link is still scarce. We aimed to examine the association between SARS-CoV-2 infection and NOAF. METHODS: We conducted a nested-case control study in a cohort of 2,931,046 adults from the largest healthcare provider in Israel. Subjects were followed from March 1st, 2020, until June 30th, 2022, for the occurrence of NOAF. Ten randomly selected controls were matched to each case of NOAF on age, sex, and duration of follow-up. Exposure to SARS-CoV-2 infection in the prior 30 days was assessed in cases and controls. To account for surveillance bias we performed a lag-time analysis and assessed the association with a negative control exposure (low back pain). Data was analyzed using conditional logistic regression. RESULTS: During the follow-up 18,981 patients developed NOAF and were matched to 189,810 controls. The mean age of cases and matched controls was 73.8 ± 13 years, and 51.1% of them were women. Multivariable analysis showed that SARS-CoV-2 infection was associated with an increased risk of NOAF; adjusted-OR, 4.24 (95% CI, 3.89-4.62). The association remained significant on lag-time analysis; however, the strength of the association was gradually attenuated with increasing lag-time but stabilized around a lag-time of 20 days. The negative control exposure (low back pain) was associated only with small increased risk of NOAF; adjusted-OR of 1.13 (95% CI, 1.02-1.26). CONCLUSION: SARS-CoV-2 infection appears to be associated with increased risk of NOAF.

Screening and Surveillance Bias in Cancer.

Surveillance bias arises when differences in the frequency of a condition are due to changes in the modality of detection rather than to a difference in the actual risk of the condition. This bias hampers the surveillance of scrutiny-dependent cancers, leading to misinterpretations of cancer trends, risk factor identification, and, consequently, to the wrong public health actions.

Exploring State Level Factors Associated with Short-Stays in Child Welfare: The Role of Systemic Risk and Surveillance.

Growing attention has been directed toward children who are placed in out-of-home care by child welfare authorities for less than 30 days, deemed "short-stayers". This exploratory study uses multiple national child welfare and population data sources to identify macro level factors associated with short-stays. Two-level logistic regression modeling was conducted to explore how state-level factors were associated with risk of short-stays. Factors associated with lower odds of short-stays included living in a state with a centralized child welfare reporting structure and with greater food insecurity. Factors associated with greater odds included living in a state with a higher percentage of the state's population enrolled in the Supplemental Nutrition Assistance Program and states with more police per capita. Multiple state level factors were associated short-stay risk, which suggests broader systemic factors contribute to these brief removals. Findings suggest greater surveillance by police and social services increases risk of short-stays, which likely have implications for child welfare policy and practice.

Prescription rates for commonly used drugs before and after a prostate cancer diagnosis.

PURPOSE: To investigate differences in prescription rates of commonly used drugs among prostate cancer patients and cancer-free comparisons and between patients diagnosed with localized and non-localized disease. METHODS: We conducted a register-based study including all men aged 50-85 years diagnosed with prostate cancer in Denmark from 1998 to 2015 and an age-matched cancer-free comparison cohort. We calculated the number of new and total prescriptions from three years before to three years after the date of diagnosis of the case for selected drug classes divided by the number of person-months and stratified by stage at diagnosis. RESULTS: We included 54,286 prostate cancer patients and 249,645 matched comparisons. 30,712 patients were diagnosed with localized disease and 12,884 with non-localized disease. The rates of new prescriptions increased considerably among patients within the year before the diagnosis. Hereafter the rates varied between drug classes. For most drug classes, total prescription rates for patients and comparisons increased similarly in the study period. Total prescription rates varied between men with localized and non-localized disease for all drug classes apart from statins. CONCLUSION: Our findings indicate that a large proportion of prostate cancer cases are likely diagnosed during medical work-up for other reasons than prostate cancer. Increased rates occur within the last year before diagnosis and future studies on the interaction between drug use and prostate cancer should at least include a one year pre-diagnostic lag-time. Post-diagnostic prescription rates demonstrated an increased use of drugs most likely associated with the consequences of the disease.

Who knew? The misleading specificity of "double-blind" and what to do about it.

BACKGROUND: In randomized trials, the term "double-blind" (and its derivatives, single- and triple-blind, fully blind, and partially blind or masked) has no standard or widely accepted definition. Agreement about which groups are blinded is poor, and authors using these terms often do not identify which groups were blinded, despite specific reporting guidelines to the contrary. Nevertheless, many readers assume-incorrectly-that they know which groups are blinded. Thus, the term is ambiguous at best, misleading at worst, and, in either case, interferes with the accurate reporting, interpretation, and evaluation of randomized trials. The problems with the terms have been thoroughly documented in the literature, and many authors have recommended that they be abandoned. PROPOSAL: We and our co-signers suggest eliminating the use of adjectives that modify "blinding" in randomized trials; a trial would be described as either blinded or unblinded. We also propose that authors report in a standard table which groups or individuals were blinded, what they were blinded to, how blinding was implemented, and whether blinding was maintained. Individuals with dual responsibilities, such as caregiving and data collecting, would also be identified. If blinding was compromised, authors should describe the potential implications of the loss of blinding on interpreting the results. CONCLUSION: "Double blind" and its derivatives are terms with little to recommend their continued use. Eliminating the use of adjectives that impart a false specificity to the term would reduce misinterpretations, and recommending that authors report who was blinded to what and how in a standard table would require them to be specific about which groups and individuals were blinded.

Disparity in child welfare referrals from public schools: An example of Simpson's Paradox?

BACKGROUND: Black children continue to be found in child welfare outcome measures at rates nearly double those of White children in the United States. Researchers have turned from bias theory to risk theory, arguing that disparity disappears when considering only the subgroup of children in poverty. In this study, we consider whether this phenomenon is an example of Simpson's Paradox, where aggregate findings are confounded by a third factor. PARTICIPANTS: We created a dataset by matching child welfare data to schools in a metropolitan California county. METHODS: We consider measures of poverty and racial-ethnic student composition as possible confounders, utilizing compositional data analysis for the latter. Traditional linear and ridge regression models were used to calculate the unadjusted and adjusted effects of each independent variable. RESULTS: We find only partial evidence of Simpson's Paradox, in that Black to White disparity only disappears in the highest quartile of poverty. Holding poverty constant, only increasing student population non-White composition was significantly associated with reducing Black to White disparity ratios. CONCLUSION: In a small, exploratory study, we find that while poverty may serve as an equalizer, diversity racial/ethnic student body composition may serve as a neutralizer. We find that underlying causes of disparity are complex and caution against endorsement of single theories to explain the disproportionate representation of Black children in child welfare. We find utility in analyzing child welfare data with concepts and techniques common in other disciplines and highlight several weaknesses of current child welfare informatics which impact both program evaluation and research.

Increased Incidence of Thyroid Cancer among World Trade Center First Responders: A Descriptive Epidemiological Assessment.

An increased incidence of thyroid cancer among 9/11 rescue workers has been reported, the etiology of which remains unclear but which may, at least partly, be the result of the increased medical surveillance this group undergoes. This study aimed to investigate thyroid cancer in World Trade Center (WTC) responders by looking at the demographic data and questionnaire responses of thyroid cancer cases from the Mount Sinai WTC Health Program (WTCHP). WTCHP thyroid cancer tumors were of a similar size (p = 0.4), and were diagnosed at a similar age (p = 0.2) compared to a subset of thyroid cancer cases treated at Mount Sinai without WTC exposure. These results do not support the surveillance bias hypothesis, under which smaller tumors are expected to be diagnosed at earlier ages. WTCHP thyroid cancer cases also reported a past history of radiation exposure and a family history of thyroid conditions at lower rates than expected, with higher than expected rates of previous cancer diagnoses, family histories of other cancers, and high Body Mass Indexes (BMIs). Further research is needed to better understand the underlying risk factors that may play a role in the development of thyroid cancer in this group.

The Patient Safety Indicator Perioperative Pulmonary Embolism or Deep Vein Thrombosis: Is there associated surveillance bias in the Veterans Health Administration?

BACKGROUND: Studies disagree whether surveillance bias is associated with perioperative venous thromboembolism (VTE) performance measures. A prior VA study used a chart-based outcome; no studies have used the fully specified administrative data-based AHRQ Patient Safety Indicator, PSI-12, as their primary outcome. If surveillance bias were present, we hypothesized that inpatient surveillance rates would be associated with higher PSI-12 rates, but with lower post-discharge VTE rates. METHODS: Using VA data, we examined Pearson correlations between hospital-level VTE imaging rates and risk-adjusted PSI-12 rates and post-discharge VTE rates. To determine the robustness of findings, we conducted several sensitivity analyses. RESULTS: Hospital imaging rates were positively correlated with both PSI-12 (r = 0.24, p = 0.01) and post-discharge VTE rates (r = 0.16, p = 0.09). Sensitivity analyses yielded similar findings. CONCLUSIONS: Like the prior VA study, we found no evidence of PSI-12-related surveillance bias. Given the use of PSI-12 in nationwide measurement, these findings warrant replication using similar methods in the non-VA setting.

Does Surveillance Bias Influence the Validity of Measures of Inpatient Complications? A Systematic Review.

Surveillance bias may threaten the accuracy of inpatient complication measures. A systematic literature review was conducted to examine whether surveillance bias influences the validity of selected Patient Safety Indicator- and health care associated infection-related measures. Ten venous thromboembolism (VTE) articles were identified: 7 trauma related, 3 postoperative, and 1 central line-associated bloodstream infection (CLABSI) article. Nine VTE articles found positive associations between deep vein thrombosis imaging and VTE diagnoses. Because imaging also may be symptom driven, most studies performed additional analyses to corroborate findings. Six trauma-related and 2 postoperative VTE studies concluded that surveillance bias was present, the latter based on circumstantial evidence. The non-VTE study found a significant positive correlation between surveillance aggressiveness and intensive care unit CLABSI rates. Even considering VTE, relatively little is known about the impact of surveillance bias on inpatient complication measures. Given the implications of misclassifying hospitals on quality, this issue requires further investigation using more direct measurement methods.

Surveillance Bias in Child Maltreatment: A Tempest in a Teapot.

Background: Children are believed to be more likely to be reported for maltreatment while they are working with mental health or social service professionals. This "surveillance bias" has been claimed to inflate reporting by fifty percent or more, and has been used to explain why interventions such as home visiting fail to reduce official maltreatment reporting rates. Methods: We use national child abuse reporting data (n = 825,763), supplemented by more detailed regional data from a multi-agency administrative data study (n = 7185). We determine the percentage of all re-reports made uniquely by mental health and social service providers within and across generations, the report sources which could be subject to surveillance bias. Results: At three years after the initial Child protective services (CPS) report, the total percentage of national reports uniquely made by mental health or social service providers is less than 10%, making it impossible that surveillance bias could massively inflate CPS reporting in this sample. Analysis of national data find evidence of a very small (+4.54%) initial surveillance bias "bump" among served cases which decays to +1.84% within three years. Our analysis of regional data showed similar or weaker effects. Conclusions: Surveillance bias effects appear to exist, but are very small.

Influenza vaccine and survival in acute heart failure.

AIMS: Influenza vaccine is a well-recommended secondary prevention measure for improving survival in patients with coronary artery disease, but it has generally been less studied in heart failure. We ask whether having influenza vaccination is associated with survival among patients with acute heart failure (HF). METHODS AND RESULTS: This was a prospective population-based cohort study accompanied by an analysis of two cross-sectional population samples for external validation of baseline characteristics differences. We analysed all 1964 ambulatory patients with acute HF aged ≥50 years who were admitted to the Heart Failure Survey in Israel (HFSIS). We used the Israel Health Survey (IHS) 2009 and the Behavioural Risk Factor Surveillance System (BRFSS) 2003-2004 surveys (274 535 participants) for external validation. In the HFSIS, the multivariate-adjusted hazard ratios for in-hospital, 1 and 4 year mortality outcomes of influenza-vaccinated patients were 0.71 (P = 0.19), 0.81 (P = 0.04), and 0.83 (P = 0.006), respectively. In the IHS validation sample, a recent physician visit [odds ratio (OR) 1.61; 95% confidence interval (CI) 1.43-1.80] or having supplementary health insurance (OR 1.39; 95% CI 1.19-1.61) were associated with higher likelihood of being vaccinated against influenza. In the BRFSS validation sample, having > 1 healthcare providers (OR 2.31; 95% CI 2.22-2.40) or having any healthcare coverage were associated with higher likelihood of being vaccinated (OR 1.59; 95% CI 1.54-1.65). CONCLUSIONS: Influenza vaccine might improve survival among patients with acute HF. This association, however, could be affected by unmeasured confounding and bias due to baseline medical surveillance and socioeconomic differences between vaccinated and non-vaccinated patients.