Estimating the pertussis burden in adolescents and adults in the United States between 2007 and 2019

ABSTRACT

We developed a machine learning algorithm to identify undiagnosed pertussis episodes in adolescent and adult patients with reported acute respiratory disease (ARD) using clinician notes in an electronic healthcare record (EHR) database. Here, we utilized the algorithm to better estimate the overall pertussis incidence within the Optum Humedica clinical repository from 1 January 2007 through 31 December 2019. The incidence of diagnosed pertussis episodes was 1–5 per 100,000 annually, consistent with data registered by the US Centers for Disease Control and Prevention (CDC) over the same time period. Among 18,573,496 ARD episodes assessed, 1,053,946 were identified (i.e. algorithm-identified) as likely undiagnosed pertussis episodes. Accounting for these undiagnosed pertussis episodes increased the estimated pertussis incidence by 110-fold on average (34–474 per 100,000 annually). Risk factors for pertussis episodes (diagnosed and algorithm-identified) included asthma (Odds ratio [OR] 2.14; 2.12–2.16), immunodeficiency (OR 1.85; 1.78–1.91), chronic obstructive pulmonary disease (OR 1.63; 1.61–1.65), obesity (OR 1.44; 1.43–1.45), Crohn’s disease (OR 1.39; 1.33–1.45), diabetes type 1 (OR 1.21; 1.17–1.24) and type 2 (OR 1.12; 1.1–1.13). Of note, all these risk factors, except Crohn’s disease, increased the likelihood of severe pertussis. In conclusion, the incidence of pertussis in the adolescent and adult population in the USA is likely substantial, but considerably under-recognized, highlighting the need for improved clinical awareness of the disease and for improved control strategies in this population. These results will help better inform public health vaccination and booster programs, particularly in those with underlying comorbidities.

KEYWORDS:

• Database
• electronic health record
• machine learning
• pertussis
• predictive modeling
• incidence
• risk factors
• comorbidities

Introduction

Pertussis (whooping cough), originally considered as a vaccine-preventable childhood disease, remains a persistent problem across all age groups. Immunity due to vaccination or previous infection wanes over time. As a result, without regular boosters, adolescents and adults are susceptible to infection and disease, and have represented a significant proportion of cases reported during outbreaks in recent years.¹ They are also recognized reservoirs of pertussis infections and may play a role in the resurgence of the disease, and in periodic epidemics.^2–4 However, ascertaining the incidence of pertussis in these populations is difficult, as the disease is generally not well recognized, diagnosed, or reported. Existing clinical case definitions are largely based on clinical presentations in young children,⁵ and are probably not appropriate for diagnosing pertussis in adolescents and adults, who often have atypical presentations.⁶ The underreporting of pertussis in these patients, including the elderly, might also be due to the perceived mild nature of the disease or not even consider as a diagnosis in otherwise healthy individuals. However, since the elderly are often affected by chronic diseases, there is concern that pertussis may lead to severe health outcomes in individuals with existing comorbidities or health conditions.⁷ Those with asthma, chronic obstructive pulmonary disease (COPD), or obesity appear to be at increased risk of pertussis disease.⁸ Indeed, asthma and/or COPD exacerbations are among the most common admission diagnoses in pertussis-related hospitalizations.⁹ Furthermore, immunodeficiency and smoking appear to be associated with worsened pertussis symptoms and increased pertussis-related hospitalization.⁸

We previously described the development of a machine learning algorithm to identify undiagnosed pertussis episodes in adolescent and adult patients with reported acute respiratory disease (ARD) using clinician notes contained within an electronic healthcare record (EHR) database.¹⁰ Here, we utilized the algorithm to identify possible or probable undiagnosed pertussis episodes recorded in a EHR database, and combined them with diagnosed pertussis episodes to have a better estimate of the overall pertussis incidence in this population. In addition, we assessed the impact of underlying comorbidities/conditions on the likelihood of pertussis or a more severe form of the disease.

Methods

Data sources

This retrospective, observational, cohort study was conducted in the Optum Humedica longitudinal clinical repository, a proprietary EHR database. Details of the Optum Humedica database have been described previously.¹⁰ This database contains EHRs for about 80 million patients across the USA with all types of insurance as well as those uninsured, across the continuum of care.

Pertussis event finding algorithm

A decision tree-based machine learning algorithm with supervised learning was deployed to identify undiagnosed pertussis episodes in adolescent and adult patients, comparing physicians’ notes in Optum Humedica from January 2007 to December 2019 between a cohort of diagnosed pertussis cases and a cohort of patients diagnosed with ARD, on the basis of cough and at least 7 other descriptive symptoms such as persistent cough, paroxysmal cough, whooping, vomiting, notably, post-tussive vomiting, and cyanosis.¹⁰ The algorithm identified whoop, whooping cough, other types of cough (especially persistent, severe, and long-lasting), and post-tussive vomiting as important predictors of pertussis. The pertussis event identification algorithm rendered a probability score of the likelihood of each ARD episode being pertussis. Balancing reduction of false positives (specificity 94%) and optimization of recall (i.e. correct identification of diagnosed pertussis episodes; sensitivity 71%) in algorithm performance evaluation, a probability score threshold of 0.5 was selected for an episode to be considered as likely undiagnosed pertussis in this study. In addition, since the algorithm ranks events by attributing probabilities, this allows for the top suspected clinical events to be assessed in priority.

Study population for incidence calculation

The cohort of diagnosed pertussis episodes occurring in individuals aged ≥11 years at index date, identified by ICD9 codes 0330, 0339, and 4843, and by ICD10 codes A3700, A3701, A3790, and A3791, and with associated physician’s notes were combined with the cohort of algorithm-identified pertussis episodes among ARD episodes occurring in individuals aged ≥11 years at index date with associated physician’s notes including at least one description of cough. This combined population formed the pertussis cohort. The population at risk denominator for incidence estimation was formed from the general Optum population aged ≥11 years with associated physician’s notes.

Study populations for risk factor identification

The pertussis episode cohort was comprised of those with diagnosed and algorithm-identified pertussis episodes included in the incidence estimation, but excluded those with missing information on vaccination or potential risk factors of interest (e.g., asthma, immunodeficiency, COPD, obesity, Crohn’s disease, diabetes mellitus types 1 and 2 [see Table S1 for ICD codes assessed]) as identified by ICD diagnosis codes. Patients with missing information or with <8 week’s coverage post-index date were also excluded. Only one (randomly selected if they had more than one episode) episode of pertussis per patient was considered. The non-pertussis cohort was constituted from ARD patients classified as negative for pertussis by the algorithm, with similar exclusions applied as for the pertussis episode cohort. For the analysis of risk factors for pertussis disease, pertussis episodes were matched 1:1 to non-pertussis ARD episodes by sex, age, and Tdap vaccination status. For the analysis of risk factors for severe pertussis, the pertussis cohort was divided into 2 sub-cohorts, those with severe episodes and non-severe episodes. Severe pertussis episodes were defined as pertussis episodes (diagnosed and algorithm-identified) presenting within 8 weeks of the index date, with at least one of the following complications identified from literature,⁵ and detected in Optum by ICD codes: encephalopathy, syncope, seizures, pneumothorax, pneumonia, rectal prolapse, urinary incontinence, rib fracture, and subdural hemorrhage (Table S2). Non-severe pertussis episodes were defined as those of the same cohort not presenting with any of the above complications in the 8 weeks following the index date. After random selection of a single episode per patient, severe pertussis episodes were matched to non-severe pertussis episodes 1:1 by sex, age, and Tdap vaccination status.

Statistical analyses

The diagnosed and estimated overall incidence of pertussis per year was calculated using the following formula:

$\mathit{Incidence}\left(\mathit{diagnosed}\right)\text{break}=\frac{\mathit{Number}\mathit{of}\mathit{patients}\mathit{with}\mathit{diagnosed}\mathit{pertussis}\mathit{episodes}}{\mathit{Number}\mathit{of}\mathit{patients}\mathit{in}\mathit{the}\mathit{general}\mathit{population}\mathit{cohort}}$

$\begin{array}{rl}& \mathit{Incidence}\left(\mathit{estimated}\mathit{overall}\right)\\ & =\frac{\begin{array}{c}\mathrm{Number}of\mathrm{patients}\mathrm{with}\\ \mathrm{diagnosed}or\mathrm{algorithm}-\mathrm{identfied}\mathrm{pertussis}\mathrm{episodes}\end{array}}{\mathit{Number}\mathit{of}\mathit{patients}\mathit{in}\mathit{the}\mathit{general}\mathit{population}\mathit{cohort}}\end{array}$

Diagnosed and estimated overall pertussis incidence was calculated for all patients and by age group (11–18 years; 19–44 years; 45–64 years; ≥65 years); these rates were compared to those reported to the US Centers for Disease Control and Prevention (CDC).

Demographic and clinical features (symptoms) of cases were summarized using descriptive statistics, and included: age (years, and age group); sex (female, male, unknown); race (Caucasian, African American, Asian, other/unknown); ethnicity: (Hispanic, not-Hispanic, other/unknown); region: (Midwest, South, Northeast, West, other/unknown); coverage (years); and individual symptoms.

To assess the impact of underlying comorbidities/conditions on the likelihood of pertussis or more severe disease, odds ratios (ORs) and their 95% confidence intervals (95% CIs) were calculated. Features were considered non-significant if the corresponding OR had a p value> .01 and were removed from the model provided this did not impact the coefficient of other features by more than 15%. If the removal of a feature from the model made the coefficients of other features vary by more than 15%, then the feature was considered a confounder and was retained in the final model. Odds ratios were calculated for all patients and by age group.

Results

Diagnosed and undiagnosed (i.e. algorithm-identified) pertussis cohorts and general population cohort

The inclusion flow chart defining the diagnosed and algorithm-identified (undiagnosed) pertussis cohorts and the general population is depicted in Figure 1. The demographic characteristics of the cases with algorithm-identified pertussis episodes were similar to those of the ARD cohort (Table 1). However, there were more adolescent and elderly patients in the diagnosed pertussis cohort compared to the algorithm-identified pertussis cohort. Most (69%) algorithm-identified pertussis episodes occurred in patients aged 19–64 years. Overall, the sex distribution was similar in all cohorts, with females outnumbering males by about 2:1.

Figure 1. The inclusion flow charts defining the pertussis diagnosed and undiagnosed cohorts, and general population.

Please use high resolution file (tiff) provided

Table 1. Demographic characteristics of the pertussis diagnosed and undiagnosed episodes and ARD episodes.

	Identified undiagnosed/(i.e., algorithm identified) pertussis episodes (0.5 threshold)		Acute respiratory disease episodes – model input		Diagnosed pertussis episodes
	Number of episodes	Share, %	Number of episodes	Share, %	Number of episodes	Share, %
Sex
Female	685871	65.08	11583310	62.36	9022	64.05
Male	368075	34.92	6990186	37.64	5052	35.87
Unknown					12	0.09
Age at ID
Mean (SD)	45 (20)		46 (21)		48 (25)
Min	11		11		11
Median (Q1-Q3)	45 (28–61)		46 (28–62)		50 (23–72)
Max	89		89		89
Age group at ID
11–18	128213	12.17	2359338	12.70	2884	20.47
19–44	397650	37.73	6519020	35.10	3302	23.44
45–64	326783	31.01	5768851	31.06	2904	20.62
>64	201300	19.10	3926287	21.14	4996	35.47
Race
African American	114768	10.89	1721077	9.27	744	5.28
Asian	18355	1.74	310840	1.67	249	1.77
Caucasian	852031	80.84	15244663	82.08	11926	84.67
Other/Unknown	68792	6.53	1296916	6.98	1167	8.28
Ethnicity
Hispanic	60077	5.70	967861	5.21	732	5.20
Not Hispanic	925456	87.81	16360523	88.09	11774	83.59
Unknown	68413	6.49	1245112	6.70	1580	11.22
Region
Midwest	543869	51.60	9473718	51.01	4887	34.69
Northeast	86687	8.22	1431360	7.71	1886	13.39
South	275513	26.14	5361384	28.87	5342	37.92
West	98514	9.35	1464343	7.88	1428	10.14
Other/Unknown	49363	4.68	842691	4.54	543	3.85

ID, index date.

Features related to cough duration and severity – such as persistent cough and severe, frequent cough – tended to be more prevalent in algorithm-identified pertussis episodes than in the ARD cohort (26% vs 3%, and 23% vs 3%, respectively; Table S3), as were symptoms and signs directly linked to the presence of an infectious agent in the lower respiratory tract, such as adventitious breath sounds, dyspnea, and chest discomfort or pain (59% vs 40%, 56% vs 33%, and 39% vs 21% in algorithm-identified pertussis episodes vs ARD episodes, respectively). Of note, the algorithm-identified pertussis episodes had lower prevalence of symptoms linked to the upper respiratory tract, such as pharyngitis (31 vs 52%) and sinus pain or pressure (20% vs 33%), than the ARD cohort.

The prevalence of symptoms indicative of pertussis was generally higher in the algorithm-identified pertussis cohort than in the diagnosed pertussis cohort. For instance, the prevalence of cough was 97% among algorithm-identified episodes, and 44% among the diagnosed pertussis episodes. This difference was largely driven by the filters applied for cohort selection during algorithm testing, which included cough in combination with at least seven other symptoms; therefore, the algorithm was by design biased toward identifying events with cough which may coincide with more “classic” symptomatic cases. Vomiting was more prevalent in patients with algorithm-identified pertussis episodes than in those with diagnosed pertussis episodes (38% and 20%, respectively), but post-tussive vomiting was reported at similar rates (6% and 5%, respectively). Whoop was rarely reported in the providers’ notes in both groups.

Estimated pertussis incidence

The incidence of diagnosed pertussis cases reported in the Optum Humedica database is summarized in Figure 2. Of note, the incidence dynamics of diagnosed pertussis episodes over time was consistent with the fluctuations in the data registered by the CDC.¹¹ However, the incidence in adolescents appears to be underestimated in the Optum Humedica database compared to the CDC data, which might be partially explained by underrepresentation of the adolescent population in the database.¹² The inclusion of algorithm-identified pertussis episodes increased the estimated pertussis incidence by 110-fold (range 34–118-fold for different age groups) on average (Figure 3). In addition, the overall estimated pertussis incidence in the database increased over time through to 2016. While the incidence trends measured by our method did not parallel those observed with CDC data or diagnosed pertussis episodes in the database, they do show steeper slopes corresponding to peaks in the CDC data (e.g., 2012). The overall estimated pertussis incidence increased over time until 2015–16 for all the age groups assessed, with the highest estimates consistently reported among adolescents (aged 11–18 years) (Figure 4a). A decrease in estimated incidence was noted in all age groups after 2015–2016, increasing again among adolescents in 2019. Estimated pertussis incidence decreased with age, as observed in the CDC data and diagnosed pertussis episodes in the database. The exception was for elderly adults (aged ≥65 years), among whom diagnosed pertussis incidence more than doubled in 2014, surpassing in subsequent years the diagnosed pertussis incidence observed for all other adult age groups (Figure 4b). This trend was not observed in the estimated incidence of pertussis.

Figure 2. Diagnosed pertussis incidence in Optum Humedica vs registered (reported) cases according to CDC.

Please use the high resolution file (tiff file) provided

Figure 3. Estimated pertussis incidence (inclusion of undiagnosed pertussis cases).

Please use the high resolution file (tiff file) provided

Figure 4. Pertussis incidence by age group: (a) Based on diagnosed and algorithm-identified (undiagnosed) pertussis episodes in Optum Humedica vs (b) Based on diagnosed pertussis episodes in Optum Humedica.

Please use the high resolution file (tiff) provided

Risk factors for pertussis disease

Among the matched cohorts (pertussis to non-pertussis episodes) the prevalence of the risk factors of interest was higher in those with pertussis episodes (Figure 5). Obesity and asthma were the most prevalent comorbidities/conditions. Prevalence of obesity was lowest in the youngest age group (13.0% and 8.5% in pertussis and non-pertussis patients, respectively) and highest in those aged 45–64 years (39.5% and 27.9%) (Table S4). Asthma prevalence decreased slightly with age in both cohorts (from 36.0% in 11–18-year-olds to 27.8% in >64-year-olds in positive pertussis episodes and from 18.7% in 11–18-year-olds to 15.6% in >64-year-olds in negative pertussis episodes). COPD was two-fold more prevalent in patients with pertussis episodes compared to those with non-pertussis episodes. As expected, COPD was rare among younger patients (0.2% and 0.1% in pertussis and non-pertussis 11–18-year-olds, respectively); the highest prevalence was in those aged ≥65 years. Type 1 diabetes prevalence also increased with age in pertussis and non-pertussis patients, though overall prevalence remained low across age groups.

Figure 5. Prevalence of underlying morbidities/conditions (after cohort matching).

Please use the high resolution file (tiff file) provided

All risk factors of interest appear to increase the likelihood of ARD being caused by pertussis, especially asthma (Figure 6). These risk factors were generally consistent across the age groups, but with a trend toward lower odds ratios with increasing age (Table S5). Asthma remained the most important risk factor for pertussis across age groups (apart from patients aged 45–64 years) and had a particularly pronounced impact on the probability of an ARD being pertussis in the younger population (odds ratios of 2.37 in patients aged 11–18 years old and 2.28 aged 19–44 years). COPD was an important risk factor of an ARD being pertussis across age groups, and the most important for patients aged 45–64 years (odds ratio 2.07).

Figure 6. Odds ratios (95% confidence interval) of underlying morbidities/conditions impacting the likelihood pertussis in adolescents and adults with ARD.

Please use the high resolution tiff file provided

Risk factors of pertussis severity

All the risk factors of interest, except Crohn’s disease, appear to increase the likelihood of severe pertussis (Figure 7). COPD was the most important risk factor for severe pertussis. There was a trend toward lower odds ratios with age (Table S6), suggesting that these have less impact on the likelihood of severe pertussis in older patients. COPD and immunodeficiency were risk factors that increased the likelihood of severe pertussis in the youngest age group in particular.

Figure 7. Odds ratios (95% confidence interval) of underlying morbidities/conditions impacting the likelihood of severe pertussis in adolescents and adults with ARD.

Please use the high resolution file (tiff file) provided

Discussion

We used a machine learning algorithm to identify likely undiagnosed pertussis episodes recorded within a population-based EHR database with the intention to better estimate the pertussis disease burden in adolescents and adults, and assess the impact of underlying comorbidities/conditions on the likelihood of pertussis disease or the development of more severe disease. The incidence of diagnosed pertussis cases within the database followed the same trend as that reported by the CDC over time. Adolescents aged 11–18 years had the highest pertussis incidence rates, also consistent with CDC data,¹¹ but incidence rates of diagnosed pertussis observed for this group in the EHR database were lower. Although this age group typically has a higher vaccination rate than adults, the higher pertussis incidence in adolescents may be the result of several factors. Adolescents may be at a higher risk of pertussis resulting from less life-time exposure to natural infection boosting of their immunity, and at the same time from higher exposure as a result of more intense social contact networks and risk of contagion. In addition, adolescents may have a higher likelihood of consultation for illness than adults resulting in a higher probability of pertussis being detected or medical records showing symptoms consistent with pertussis. The inclusion of algorithm-identified episodes increased the estimated pertussis incidence by 110-fold on average, highlighting the extent of potential underreporting and/or lack of recognition of the disease in the population assessed. Interestingly, females outnumbered males by about 2:1, though this might be due to a bias linked to the database used. The distribution by race group was also modified by algorithm identified pertussis episodes with increased representation of African-Americans, more consistent with the proportion of this race group in the general US population.¹³ While the proportion of diagnosed pertussis was consistent with a previous CDC analysis,⁹ the consistency between CDC and Optum in lower representation of African-Americans among diagnosed pertussis than in the general population suggests a need to investigate further factors affecting the risk of pertussis and the probability of diagnosis in this race group compared to others. We also found that patients with ARD across all groups with comorbidities/conditions such as asthma, immunodeficiency, COPD, obesity, Crohn’s disease, and diabetes (type 1 and type 2) were at a higher risk of pertussis. These comorbidities/conditions, except Crohn’s disease, also increased the likelihood of severe pertussis.

Our study suggests that pertussis in adolescents and adults occurs more frequently than currently reported to health authorities or identified in practice. Underreporting of pertussis cases in adults is well recognized both in older adults aged ≥50 years,^14,15 as well as those aged <50 years.¹⁶ Another database analysis suggested that, assuming that a fraction of all cough illness were statistically attributable to pertussis, the estimated pertussis incidences in adults aged ≥50 years in the USA were 42–105-times higher depending on the year (2006–2010) than medically attended cases.¹⁵ The estimated incidences of cough illness attributed to B. pertussis during the study period were on average 202 and 257 per 100,000 among people aged 50–64 and ≥65 years, respectively, and increased over the years in both age groups.¹⁵ In another similar database analysis, again assuming that a fraction of all cough illness were statistically attributable to pertussis, the annual incidence of pertussis was estimated at approximately 58–93-times higher than diagnosed pertussis in those aged <50 years (defined as claims for pertussis identified with ICD codes) depending on the year (2008–2013).¹⁶ The consistency of our results, showing 34- to 118-fold (mean 110-fold) higher incidence of pertussis than diagnosed, with these previous estimates using a different methodology provides confidence in the validity of our approach and results. Our estimates of incidence were generally higher than these previous results, but our study period also covered years in which the US experienced large outbreaks with continued heightened incidence of pertussis.

We showed the following comorbidities/conditions increased the likelihood of pertussis, consistent with other studies: asthma,^17–19 COPD,^18,20 and obesity.¹⁹ In addition, these comorbidities/conditions, except Crohn’s disease, appeared to increase the likelihood of severe pertussis.⁸ To the best of our knowledge, our analysis is the first to more definitively show that immunodeficiency, Crohn’s disease, and diabetes (type 1 and type 2) also increase the likelihood of pertussis. Interestingly, both obesity and type 2 diabetes are associated with insulin resistance,²¹ and it is possible these individuals may be more prone to infectious agents since chronic low-grade inflammation may contribute to the development of these two conditions.^22,23 Those with Crohn’s disease may be at increased risk of infection due to the use of immunosuppressives in the management of the condition,²⁴ which also impair immunogenicity to pertussis vaccination.²⁵ Indeed, many patients with inflammatory bowel disease appear to lack detectable immunity to pertussis.²⁶ The severity of Crohn’s disease and the drugs used in its management may have been a confounder for this risk factor in our analyses as it was previously found that milder forms of Crohn’s disease for which non-immunosuppressive treatments (e.g. aminosalicylates) may be prescribed did not increase the risk of infection.²⁷ Our observation that Crohn’s disease did not increase the risk of severe pertussis support this hypothesis, but as our study did not access treatment data, we were not able to investigate this further. A previous study reported high rates of asthma, diabetes, immunocompromising conditions, COPD, and obesity among hospitalized pertussis cases aged≥12 years,⁹ which suggests that these could be potential risk factors for severe pertussis requiring hospitalization. There was also a high prevalence of respiratory viral co-infection among patients hospitalized with pertussis. It is also possible that pertussis may exacerbate underlying comorbidities/conditions. Indeed, pertussis appeared to play a role in acute exacerbations of chronic bronchitis.²⁸ Among asthma patients, those with pertussis have been shown to have more asthma symptoms and airway obstruction than those without pertussis.²⁹ We observed a trend toward lower odds ratios for all risk factors assessed with increasing age. It is possible that age itself may be a risk factor influencing the severity of pertussis disease, therefore masking the effect of co-morbidities. It would therefore be informative to assess the influence of age on the severity of pertussis in future studies.

Our study has a number of limitations which have been discussed in detail previously.¹⁰ Here, although we observed that the pertussis incidence increased with time, this may in part be due to changes in data recording in the database. For example, it is possible that the symptom descriptions may be getting more comprehensive or better recorded with time leading to a higher probability of predicting an episode as pertussis by the algorithm when there are a higher number of clinical characteristics described. However, this may also be considered as a strength for future detection because algorithm training would be on rich descriptive data before applying on a prospective EHR database. It is also possible that physician awareness might have contributed to the sustained higher numbers of pertussis cases (with ICD codes) reported in the later part of the study period compared to the first few years assessed.³⁰ To better describe the episodes of pertussis and of ARDs in terms of the associated symptoms, additional data might also be used, such as the information on symptoms codified with ICD codes by the healthcare professionals. Furthermore, other features such as physician specialty, recurrent visits due to persistent cough, medications, and procedures could be valuable in increasing algorithm performance. The strength of our study includes the use of a large geographically diverse US population-based sample with comprehensive real-world data representative of typical clinical practice.

In conclusion, our study suggests that the incidence of pertussis in the adolescent and adult population in the USA is likely substantial, but considerably under-recognized, highlighting the need for improved clinical awareness of the disease and for improved control strategies in this population. These results will help to better inform the public health vaccination and booster programs.

Author contributions

CD, DM, SE, PH, MD, and MM contributed to the design of the methods. SM, SE, and PH contributed to analysis of the data. DM, SM, MD, and MM contributed to the interpretation of data. All authors were involved in drafting the article and final approval of the article. All authors are accountable for the accuracy and integrity of this report.

Role of the funding source

Sanofi was involved in the study design, accessing the electronic healthcare records database, analysis, and interpretation of data, the writing of the report; and in the decision to submit the paper for publication. All authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Acknowledgments

Editorial assistance with the preparation of the manuscript was provided by Richard Glover, inScience Communications, Springer Healthcare Ltd, Chester, UK, and was funded by Sanofi. The authors also thank Roopsha Brahma, PhD, for editorial assistance and manuscript coordination on behalf of Sanofi.

Disclosure statement

DM, SM, and CD are employees of Sanofi and may hold shares and/or stock options in the company. MD, SE, PH and MM are employees of Quinten Health who were contracted by Sanofi to conduct this research.

Supplementary data

Supplemental data for this article can be accessed on the publisher’s website at https://doi.org/10.1080/21645515.2023.2208514.

Additional information

Funding

This study was funded and sponsored by Sanofi.