Abstract
Background
Opportunistic bacterial infection is a hallmark of cystic fibrosis (CF) lung disease and early mortality. Poorly characterized prevalence changes have accompanied two decades of health improvements, with CFTR modulators likely to further affect infection epidemiology.
Methods
Bacterial prevalence change trends across birth cohorts were assessed with linear regression using 2001–2019 US CF Foundation Patient Registry data. Informative missingness was assessed, as was age-to-age infection status.
Results
Bacterial prevalence constantly changed from 2001 through 2019, with changes differing across birth cohorts. Informative censoring affected prevalence change for some organisms. Age-to-age infection status changes were greater than net changes in bacterial prevalence and varied by age.
Conclusions
CF infection epidemiology changed over two decades and will continue to do so. Understanding how modulators affect infection epidemiology will require creative designs for longitudinal prevalence change studies emphasizing prevalence changes independent of effects on lung biology.
Keywords: cystic fibrosis, infection prevalence, opportunistic infection
Forward
Complex opportunistic polymicrobial airways infection is a defining characteristic of cystic fibrosis (CF) [1]. From birth, CF airways are prone to infection with microbial opportunists, with successive (and often more concerning) bacterial species appearing in respiratory secretions over (usually shortened) lifetimes. Loss of lung function secondary to local inflammation fueled by these opportunists [2] accounts for most US CF deaths [3].
CF management has dramatically changed over the past two decades, accompanied by substantial improvements in CF population health [4]. To the extent that bacterial opportunism is a reflection of host health [5], it follows that incremental improvements in CF population health and survival have been accompanied by changes in CF airway infection prevalence [6–9] (particularly as some management practices are aimed directly at changing infection status via antimicrobial treatment [10]).
As access to CFTR modulators has expanded, there has been discussion of how modulators will change the epidemiology and natural history of CF airway infection (e.g., [11–13]), both by changing airway biology as well as access to and quality of respiratory secretions with which to assess infection. These are important and interesting questions that in some respect presume a good understanding of premodulator airway infection epidemiology and that tend to minimize previous reports of consistent epidemiologic change [6–9]. If (when) modulator treatments begin to change airway infection natural history, this change will not occur in a static system, but one which has already been changing for decades.
Interestingly, despite the association of CF airway infection and lung disease progression, our appreciation of how CF airway infection epidemiology has changed over past decades remains incomplete, at best. Most CF clinicians are familiar with cross-sectional age-specific infection prevalence data such as the plot entitled “Prevalence of Respiratory Microorganisms by Age Cohort” published in the US CF Foundation Patient Registry (CFFPR; [14]) annual report. These data rich plots (e.g., Figure 1A) are a composite of prevalence histograms for different organisms (Figure 1B highlights the Pseudomonas aeruginosa prevalence data embedded in Figure 1A) that describe the airway infection status of the CF population during a given year, but which don’t lend themselves well to studying infection prevalence changes over time. Longitudinal analyses of CF airways infection prevalence from 1995 through 2005 [6] and 2006 through 2012 CFFPR data [7], as well as from 2001 through 2016 for European CF Registry data [9], have reported annual changes in prevalence between age groups, but not from age-to-age. A single-center CF study from Australia reported changes in adult bacterial prevalence over four-year intervals [8]. However, information on age-to-age changes in airway infection epidemiology, as well as year-to-year changes in individual infection status, remains sparse.
Figure 1. P. aeruginosa prevalence from the 2019 CFFPR [3].
A; cross-sectional prevalence plot published by the CFF that describes the proportion of patients for whom each of six bacterial species (MRSA is a subcategory of Staphylococcus aureus) was detected in culture at least once during 2019 among patients having had a respiratory culture. Prevalence values for different age groups are linked by single lines of different colors. Reprinted with permission. B; plot of P. aeruginosa prevalence from panel A as a histogram. C; histogram from panel B replotted with a linear x-axis for age. Vertical dashed lines mark age categories used in panels A and B. D; Histogram of patients providing respiratory cultures in the 2019 CFFPR.
In this perspective, we show how repurposing/refitting currently available annualized CFFPR infection prevalence data collected from 2001 through 2019 can enrich our understanding of past changes in CF infection epidemiology both at the individual and cohort level. These approaches may prove useful to investigators studying effects of modulator treatments on CF infection epidemiology.
Infection prevalence by age
A natural inclination when considering Figure 1A, in which patients of different ages represent different birth cohorts, is to infer that the current bacterial prevalence of an older cohort approximates the future bacterial prevalence of a younger cohort. We understand that this approximation is not an actual longitudinal prevalence difference, as it is derived from the prevalence of different cohorts in the same year as opposed to the same cohort in different years. Age categorization, in which older birth cohorts are binned into successively larger groups (Figures 1C and D), complicates this imputation approach. More importantly, substantial error is introduced when prevalence from an older birth cohort is used to predict a younger cohort’s future prevalence, arising from an assumption that bacterial prevalence at a given age is relatively constant across successive birth cohorts. In fact, age-specific Pseudomonas aeruginosa prevalence varied broadly from 2001 through 2019 (Figure 2A) and commonly differed between adjacent birth cohorts (Figure 2B; Supplemental Figure S1), defined as the year in which the individual first appeared in the prevalence data minus their truncated age (in years) at the end of that year. Further, and consistent with incremental health improvement, age-specific P. aeruginosa prevalence for children and young adults tended to decrease across the observation period (Supplemental Figure S1), meaning that use of an older cohort to estimate future P. aeruginosa prevalence in a younger cohort would almost always produce an inflated future prevalence estimate. For example, using age-specific P. aeruginosa prevalence from the 1989 birth cohort (Figure 2C, yellow markers) to estimate age-specific P. aeruginosa prevalence for the 1990 birth cohort (red markers) results in 16 of 18 prevalence estimates being inflated, some notably, compared to what was actually observed. In Figure 2C, the annual 2009 P. aeruginosa prevalence values for 19-year-olds (from the 1990 birth cohort) and 20-year-olds (from the 1989 birth cohort) were 61.6% and 65.5%, respectively, a difference of 3.9%. However, the 2010 annual P. aeruginosa prevalence for 20-year-olds (the 1990 birth cohort) was 62.3%, an increase of only 0.7% from the prior year (as opposed to the ~3.9% prevalence increase estimated using the 2009 prevalence observed among 20-year-olds).
Figure 2. Annual P. aeruginosa prevalence by age.
A; Annual CFFPR P. aeruginosa prevalence by age from 2001 through 2019 stratified by reporting year. Data from the 2001 report (green) and 2019 report (magenta) are highlighted, Prevalence data from 2002 through 2018 reporting years are plotted in a dark blue to light blue pallet. B; Data from panel A stratified by birth cohort for 1970 through 2019 birth cohorts, with prevalence within each birth cohort connected with lines. The 1989 and 1990 cohort data are highlighted in yellow and red, respectively. C; Simple linear regression trend lines derived from 1989 (yellow) and 1990 (red) birth cohort prevalence measures from panel B. Note that axes are expanded compared to other panels. D; Trend lines as in panel C plotted for 1970 through 2010 birth cohorts, with 1989 and 1990 birth cohorts highlighted in yellow and red, respectively.
Longitudinal prevalence change
Variability in age-specific P. aeruginosa prevalence observed from 2001 through 2019 (Figure 2A; Supplemental Figure S1) suggests that year-to-year prevalence changes at a given age might be better characterized first within birth cohorts and then compared across cohorts. Stratifying prevalence estimates by birth cohorts highlights prevalence changes that occurred within cohorts during the period (Figure 2B); simple least-squares linear regressions among birth cohorts highlight differences in prevalence change trends between cohorts during the period (Figures 2C and 2D). Characterizing cohort change trends in this manner is imperfect, as linear regression only poorly represents observed trends in some cohorts (e.g., Figure 2C), but is an improvement over use of cross-sectional data to infer prevalence change (which tends to generate inflated change estimates).
Annual P. aeruginosa infection prevalence trends from 2001 through 2019 differed substantially across birth cohorts, with differences between adjacent cohorts more incremental (Figure 2D). Although P. aeruginosa prevalence at a given age decreased >1% per cohort year in children and young adults among successive birth cohorts (Figure 2B; Supplemental Figure S1), annual P. aeruginosa prevalence trends increased during the period for the 1986 through 2010 birth cohorts (Figure 2D; Supplemental Figure S2). Note that the +0.6% prevalence per registry year trend observed for the 1990 birth cohort (Supplemental Figure S2) approximates the +0.7% prevalence change observed for this cohort from age 19 to 20 (Figure 2C). In contrast, P. aeruginosa prevalence did not increase materially among older birth cohorts (having 2001 P. aeruginosa prevalence >70%) during the period.
Interestingly, there was a strong inverse correlation (R2=0.93) between a birth cohort’s annual P. aeruginosa prevalence in 2001 and the slope of that cohort’s prevalence trend line from 2001 through 2019 (Supplemental Figure S3), with (younger) birth cohorts having lower 2001 P. aeruginosa prevalence showing more rapid prevalence increases in subsequent years. Some of this may reflect regression to the mean, but it also suggests that the lower P. aeruginosa infection prevalence observed among younger birth cohorts was not accompanied by a corresponding reduction in risk of future infection.
Missing data
Annual CFFPR bacterial prevalence data are derived from registry participants who have had at least one culture result reported during the year, with each birth cohort’s contributors varying from year to year. In addition to those patients lacking culture results for a given year, “new” patients can enter birth cohorts later in life through relocation or recent CF diagnosis, while others can leave cohorts due to relocation, death, or loss to follow-up. Because of this variation, some patients are more consistently represented in the CFFPR from 2001 through 2019 than others. There are reasons to believe that bacterial prevalence “missingness” might be informative (i.e., that it may not be entirely random and may affect observed prevalence). For instance, patients diagnosed with CF as adults are likely to have a less aggressive form of CF than their birth cohort peers who were diagnosed as infants: if they are at relatively reduced risk for opportunistic infection, their entry into a cohort might reduce that cohort’s bacterial prevalence. Similarly, those patients who have been lost to follow-up due to death may have had a more aggressive form of CF and been at greater risk of opportunistic infection than their surviving peers. Their absence in later years of the period might also be associated with a reduced bacterial prevalence.
In this analysis, 44,306 unique patients were followed with respect to bacterial prevalence in the CFFPR from 2001 through 2019. Of these, only 18,381 (41.2%) were identified as being “data rich” patients (missing data from no more than one of all possible years across the observation period). Missingness was strongly associated with birth cohort, with older cohorts having a progressively lower proportion of patients being data rich. P. aeruginosa prevalence trend lines derived from data rich cohort subsets differed from trend lines derived from entire cohorts (Figure 3). Trend differences were most striking in older birth cohorts with greater missingness, where prevalence was relatively constant among data rich patients (Figure 3, black trend lines) but decreased markedly over the period for entire cohorts (Figure 3, gray trend lines), suggesting selective loss to follow-up of P. aeruginosa-infected patients, entry of P. aeruginosa-free patients, or both.
Figure 3. P. aeruginosa prevalence trends for 1970 through 2010 CFFPR birth cohorts by data richness.
Simple linear prevalence by age regressions stratified by birth cohort for all patients followed (gray lines) and limited to data-rich patients missing culture data from no more than one of all possible years from 2001 through 2019 (black lines).
Thus, although age-specific P. aeruginosa prevalence increased in adolescence and then decreased in adulthood among CFFPR patients from 2001 through 2019, prevalence declines in adulthood (among the 1970 to 1984 birth cohorts) appear to have been largely driven by informative censoring (i.e., missingness). In contrast, more dramatic trends of decreasing Staphylococcus aureus and Haemophilus influenzae prevalence within cohorts observed among data-rich patients across the period (Supplemental Figure S4) suggest that these prevalence reductions cannot be attributed to missingness. To the extent that we are interested in longitudinal infection prevalence as a measure of infection risk in people with CF, including study of prevalence trends in data-rich subgroups is warranted.
Individual culture status and status change
Up to this point, our focus has been on bacterial prevalence and prevalence change at the birth cohort level, particularly among those cohort members with complete or nearly complete data available from 2001 through 2019. However, cohort prevalence and prevalence change are of little use in the management of individuals with CF, who are either infected with a given species or are not. Although we can infer that a predicted increase in infection prevalence from 50% this year to 55% next year equates to 1 in 10 currently culture-negative cohort members converting to culture-positivity, those odds represent the risk of conversion to culture positivity in the absence of any culture-positive patients reverting to culture-negativity during the year. Indeed, all we can confidently conclude from a 50% to 55% net change in cohort prevalence among patients with data available in both years is that a) more individuals will convert to culture-positivity than will revert to negativity during the next year, with a net difference (conversion minus reversion) equal to 5% of the cohort population and b) that the odds of conversion to culture positivity among culture-negative patients are at least (but very likely exceed) 1 in 10.
Age-to-age infection analyses accounting for each patient’s infection status (yes/no) and their status from one year to the next (same/different) provide a more granular view of age-specific changes in bacterial prevalence. This approach revealed that many more patients changed infection status from age to age than were reflected by annual population prevalence changes. For instance, from 2001 through 2019, >20% of CFFPR patients changed P. aeruginosa infection status each year from birth through age 10 years. At each age beyond 10 years, more than 10% of patients with data present in consecutive years changed P. aeruginosa infection status (either changing from negative to positive or from positive to negative) from one year to the next (Figure 4A). In contrast, age-to-age change in net P. aeruginosa infection prevalence exceeded 5% only once (increasing >11% from age 0 to 1 year) and was commonly <2% through age 50 (Figure 4B).
Figure 4. Age-to-age P. aeruginosa culture status.
For each patient who had annual CFFPR culture results in any two consecutive years, annual culture results from the earlier year were compared to those from the later year and plotted by patient age at the end of the earlier year. A; proportions of patients who converted from culture-negative to -positive (black bars) and reverted from culture-positive to -negative (gray bars). B; net age-to-age change in P. aeruginosa population prevalence for the patients represented in panel A
Advantages of this age-to-age approach for assessing longitudinal prevalence change include a) it is not directly confounded by data missingness because patients require culture results in consecutive years to be included and b) it is a much less stringent approach for censoring missingness than the trend analyses above: 41,183 of 44,306 registry patients in these analyses (93%) had at least one dyad of years with consecutive culture data, versus only 41% of patients who met the data rich definition for the trend analyses in Figure 3 above.
Differences between individual age-to-age infection status change rates and net population prevalence change rates from 2001 through 2019 were also observed for other CF opportunists (Supplemental Figure S5). As with prevalence change trends, the nature of the discordance between ageto-age infection status change and net prevalence change differed by CF opportunist.
Bias and interpretation
Although we think of annualized bacterial prevalence data as reflective of lung biology and environmental exposure of the CF population, annualized prevalence is the product of multiple steps between airway biology and registry analysis that can introduce bias and artifact into infection prevalence estimation (Figure 5). Consider that beyond the presence of a bacterial species in the airway, a person with CF must visit a clinic, have a respiratory specimen collected, have that specimen cultured in a laboratory, have that culture result reported back to the clinic, and have that clinic forward the result to the registry in order to be considered “culture positive.” Inconsistencies or changes in any of these steps can confound species prevalence measures. Reasons for and frequency of an individual’s visits to their CF clinic, frequency of CF clinician culture requests [15], types of sample(s) collected [16], sampler’s technique, and culture technique may all influence the probability of an organism that is actually present in a person’s airways being correctly identified in registry prevalence data. In addition, clinical interventions introduced in response to previous culture results (e.g., acute or chronic antimicrobial interventions or changes in sampling methodology or frequency) may influence subsequent organism detection. As an example, individuals with history of lung transplant are not excluded from annual opportunistic infection prevalence data. The fundamental airways change associated with transplant, as well as a potential for upper airway re-infection of the graft, chronic administration of immunosuppressive therapies, and considerable differences in pre- versus post-transplant airway infection surveillance, all suggest that prevalence should be affected by transplant.
Figure 5. Factors that may affect observed infection prevalence and prevalence change in people with CF.
A series of intervening steps lie between CF airways that are prone to infection (upper left panel) and estimation of infection prevalence in the CF population (lower left panel). Variation introduced in intermediate steps (e.g., differences/changes in patient behavior, clinical standards of care, airway secretion sampling methodology/technique, detection methodology/technique, and data communication to registries, and analytical methodology) may all affect infection prevalence estimates.
Dependence of observed bacterial prevalence on established surveillance/reporting systems was illustrated during the COVID-19 pandemic. Between 2019 and 2020 Registry years, experienced CF respiratory clinicians were shifted to COVID coverage, CF clinic visits fell from an average of >4.5/patient-year to 2.5/patient-year, average bacterial cultures fell from 3.8/patient-year 2.2/patient-year, and P. aeruginosa prevalence for the whole population fell from 43.2% in 2019 to 32.0% in 2020 [17]. There are no biological bases for this degree of reduction in P. aeruginosa infection prevalence across the population in a single year, suggesting that changes in care interface and sampling (Figure 5) accounted for at least some of this reduction.
These will be important considerations for future studies of the effects of modulators on airways infection prevalence, as clinical improvements associated with modulator treatment have the potential to meaningfully affect downstream confounders of infection detection in unpredictable ways. For instance, modulator patients who feel better may be less prone to frequent sampling, their samples may be more likely to be throat swabs rather than sputum or lavage fluid when compared to sampling before modulators, and they may be less likely to be prescribed chronic inhaled antimicrobials for identified infections. Clinicians managing these patients may change their infection surveillance protocols precisely because they believe that their infections will be harder to identify, and a perceived change in an individual’s infection status may alter both their environmental exposures and the exposures of other patients in clinics that cohort patients based on infection status. All of these changes have the potential to affect “observed” prevalence in people with CF taking modulators in a manner unrelated to modulator-associated changes in underlying lung biology and risk of opportunistic infection.
Conclusions
CF airway infection epidemiology and natural history have continuously changed over the past twenty years as access to interventions has broadened and management practices have evolved. Our current understanding of past changes in infection epidemiology and our appreciation of how observed changes in infection prevalence may have been confounded by changes in surveillance methodology and clinical practice (as opposed to changes in lung biology and environmental exposure) are rudimentary. If we hope to understand how CFTR-modulator associated changes in lung biology will affect risk of opportunistic CF airway infection in the future, creative analytical methods and greater recognition of the role confounders play in estimating infection prevalence will be needed.
Supplementary Material
Highlights.
CF airways are prone to opportunistic infections that can drive local inflammation.
Prevalence changes were studied from 2001–2019 for 7 CF bacterial opportunists.
Overall opportunist prevalence generally fell as population health improved.
Year-to-year prevalence change rates differed across organisms and birth cohorts.
More people changed infection status yearly than was suggested by prevalence change.
Acknowledgements
The authors thank the Cystic Fibrosis Foundation for the use of CF Foundation Patient Registry data to conduct this study. Additionally, we thank the patients, care providers, and clinic coordinators at CF centers throughout the United States for their contributions to the CF Foundation Patient Registry. Finally, we thank Travis C. VanKrause for assistance with R programming.
Competing Interests
The authors have no competing interests relative to this content to report. MWK received salary support to his institution from the Cystic Fibrosis Foundation and the National Institutes of Health (UL1TR002548). JJL received salary support from the Cystic Fibrosis Foundation (CAVERL22AB0).
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
CONFLICT OF INTEREST STATEMENT
The authors have no ethical or financial conflicts that would affect the content of this submission
CRediT Statement
DRV conceived these analyses, curated and analyzed data, and drafted the original manuscript and figures. JLL contributed to the scope of analyses and to the writing and editing of the manuscript. MWK acquired the data and contributed to the scope of analyses as well as the writing/editing of the manuscript.
References
- [1].LiPuma JJ. The changing microbial epidemiology in cystic fibrosis. Clin Microbiol Rev. 2010. Apr;23(2):299–323. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [2].Ratjen F, Döring G. Cystic fibrosis. Lancet. 2003. Feb 22;361(9358):681–9. [DOI] [PubMed] [Google Scholar]
- [3].Cystic Fibrosis Foundation Patient Registry 2019 Annual Data Report. Bethesda, Maryland. ©2020 Cystic Fibrosis Foundation. [Google Scholar]
- [4].Ratjen F, VanDevanter DR. Retracing changes in cystic fibrosis understanding and management over the past twenty years. J Cyst Fibros. 2022. Jan;21(1):3–9. [DOI] [PubMed] [Google Scholar]
- [5].Casanova JL. From second thoughts on the germ theory to a full-blown host theory. Proc Natl Acad Sci U S A. 2023. Jun 27;120(26):e2301186120. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [6].Razvi S, Quittell L, Sewall A, Quinton H, Marshall B, Saiman L. Respiratory microbiology of patients with cystic fibrosis in the United States, 1995 to 2005. Chest. 2009. Dec;136(6):1554–1560. [DOI] [PubMed] [Google Scholar]
- [7].Salsgiver EL, Fink AK, Knapp EA, LiPuma JJ, Olivier KN, Marshall BC, Saiman L. Changing Epidemiology of the Respiratory Bacteriology of Patients with Cystic Fibrosis. Chest. 2016. Feb;149(2):390–400. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [8].Ramsay KA, Sandhu H, Geake JB, Ballard E, O’Rourke P, Wainwright CE, Reid DW, Kidd TJ, Bell SC. The changing prevalence of pulmonary infection in adults with cystic fibrosis: A longitudinal analysis. J Cyst Fibros. 2017. Jan;16(1):70–77. [DOI] [PubMed] [Google Scholar]
- [9].Hatziagorou E, Orenti A, Drevinek P, Kashirskaya N, Mei-Zahav M, De Boeck K; ECFSPR. Electronic address: ECFS-Patient.Registry@uz.kuleuven.ac.be; ECFSPR. Changing epidemiology of the respiratory bacteriology of patients with cystic fibrosis-data from the European cystic fibrosis society patient registry. J Cyst Fibros. 2020. May;19(3):376–383. [DOI] [PubMed] [Google Scholar]
- [10].Langton Hewer SC, Smyth AR. Antibiotic strategies for eradicating Pseudomonas aeruginosa in people with cystic fibrosis. Cochrane Database Syst Rev. 2017. Apr 25;4(4):CD004197. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [11].Rogers GB, Taylor SL, Hoffman LR, Burr LD. The impact of CFTR modulator therapies on CF airway microbiology. J Cyst Fibros. 2020. May;19(3):359–364. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [12].Harvey C, Weldon S, Elborn S, Downey DG, Taggart C. The Effect of CFTR Modulators on Airway Infection in Cystic Fibrosis. Int J Mol Sci. 2022. Mar 23;23(7):3513. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [13].Saluzzo F, Riberi L, Messore B, Loré NI, Esposito I, Bignamini E, De Rose V. CFTR Modulator Therapies: Potential Impact on Airway Infections in Cystic Fibrosis. Cells. 2022. Apr 6;11(7):1243. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14].Knapp EA, Fink AK, Goss CH, Sewall A, Ostrenga J, Dowd C, Elbert A, Petren KM, Marshall BC. The Cystic Fibrosis Foundation Patient Registry. Design and Methods of a National Observational Disease Registry. Ann Am Thorac Soc. 2016. Jul;13(7):1173–9. [DOI] [PubMed] [Google Scholar]
- [15].Shreve MR, Butler S, Kaplowitz HJ, Rabin HR, Stokes D, Light M, Regelmann WE. Impact of microbiology practice on cumulative prevalence of respiratory tract bacteria in patients with cystic fibrosis. J Clin Microbiol. 1999. Mar;37(3):753–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16].Rosenfeld M, Emerson J, Accurso F, Armstrong D, Castile R, Grimwood K, Hiatt P, McCoy K, McNamara S, Ramsey B, Wagener J. Diagnostic accuracy of oropharyngeal cultures in infants and young children with cystic fibrosis. Pediatr Pulmonol. 1999. Nov;28(5):321–8. [DOI] [PubMed] [Google Scholar]
- [17].Cystic Fibrosis Foundation Patient Registry 2020 Annual Data Report. Bethesda, Maryland. ©2021 Cystic Fibrosis Foundation. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.