Factors indicating intention to vaccinate with a COVID-19 vaccine among older U.S. adults

Abstract

Background: The success of vaccination efforts to curb the COVID-19 pandemic will require broad public uptake of immunization and highlights the importance of understanding factors associated with willingness to receive a vaccine.

Methods: U.S. adults aged 65 and older enrolled in the HeartlineTM clinical study were invited to complete a COVID-19 vaccine assessment through the HeartlineTM mobile application between November 6-20, 2020. Factors associated with willingness to receive a COVID-19 vaccine were evaluated using an ordered logistic regression as well as a Random Forest classification algorithm.

Results: Among 9,106 study participants, 81.3% (n = 7402) responded and had available demographic data. The majority (91.3%) reported a willingness to be vaccinated. Factors most strongly associated with vaccine willingness were beliefs about the safety and efficacy of COVID-19 vaccines and vaccines in general. Women and Black or African American respondents reported lower willingness to vaccinate. Among those less willing to get vaccinated, 66.2% said that they would talk with their health provider before making a decision. During the study, positive results from the first COVID-19 vaccine outcome study were released; vaccine willingness increased after this report.

Conclusions: Even among older adults at high-risk for COVID-19 complications who are participating in a longitudinal clinical study, 1 in 11 reported lack of willingness to receive COVID-19 vaccine in November 2020. Variability in vaccine willingness by gender, race, education, and income suggests the potential for uneven vaccine uptake. Education by health providers directed toward assuaging concerns about vaccine safety and efficacy can help improve vaccine acceptance among those less willing.

Trial registration: Clinicaltrials.gov NCT04276441.

Conflict of interest statement

I have read the journal’s policy and the authors of this manuscript have the following competing interests: JN, MK, GJW, JW, TS, JS, PB, RZ, NE, JT, RJ are employees of Janssen and Johnson and Johnson. JN, MK, GJW, JW, TS, JS, PB, RZ, NE, JT, RJ are employees of Janssen and Johnson and Johnson. This does not alter our adherence to PLOS ONE policies on sharing data and materials. JS reports personal fees from Amgen, personal fees from Bayer, personal fees from Merck, personal fees from Novartis, personal fees from Janssen, personal fees from Myokardia, personal fees from Blue Cross Blue Shield of Kansas City, outside the submitted work. In addition, Dr. Spertus has a patent Copyright to the KCCQ with royalties paid and Equity in Health Outcomes Sciences. This does not alter my adherence to PLOS ONE policies on sharing data and materials. MT reports grants from Janssen Inc, personal fees from Medtronic Inc, personal fees from Abbott, grants from Boehringer lngelheim, grants and personal fees from Cardiva Medical, personal fees from iRhythm, grants from Bristol Myers Squibb, grants from American Heart Association, grants from SentreHeart, personal fees from Novartis, personal fees from Biotronik, personal fees from Sanofi, personal fees from Pfizer, grants from Apple, grants from Bayer, personal fees from Myokardia, personal fees from Johnson & Johnson, personal fees from Milestone Pharmaceuticals, outside the submitted work; and Dr. Turakhia is an editor for JAMA Cardiology. This does not alter my adherence to PLOS ONE policies on sharing data and materials. LS reports personal fees from J&J, outside the submitted work. This does not alter my adherence to PLOS ONE policies on sharing data and materials. MG receives research grant support from Janssen. This does not alter my adherence to PLOS ONE policies on sharing data and materials. AMN receives consulting fees from Janssen for serving on the steering committee of the Heartline study. In addition, Dr. Navar receives funding for research to her institution from Amgen and Janssen, and honoraria and consulting fees from Amarin, Amgen, Astra Zeneca, BI, Esperion, Lilly, Sanofi, Regeneron, NovoNordisk, Novartis, The Medicines Company, New Amsterdam, Cerner, 89Bio, and Pfizer, outside the scope of this work. This does not alter my adherence to PLOS ONE policies on sharing data and materials.

Figures

Fig 1. Forest plot of willingness to vaccinate by demographics.

Shown are Odds Ratios (95% CI) for willingness to vaccinate for the different demographic characteristics. Odds Ratios were calculated using ordered logistic regression model with the 4 levels of willingness to be vaccinated as the outcome while adjusting for gender and race. Reference for each category is indicated by an open circle. na indicate not sufficient subjects for this category. ‘Native Hawaiian or Other Pacific Islander’, ‘Two or more’, and ‘Prefer not to answer’ are combined in ‘Other’.

Fig 2. Forest plot of willingness to vaccinate by survey response.

Shown are Odds Ratios (95% CI) for willingness to vaccinate for the different demographic characteristics. Odds Ratios were calculated using ordered logistic regression model with the 4 levels of willingness to be vaccinated as the outcome while adjusting for gender and race. Reference for each survey question is the option ‘neutral’ and is indicated by an open circle. For example subjects who agreed with the question ‘I am comfortable taking a COVID-19 vaccine that has short term side effects such as stomach pain or nausea if the vaccine efficiently prevents COVID-19.’ are 3.0 times more likely to be more willing as compared to those who selected ‘neutral’.

Fig 3. Result of recursive feature elimination algorithms.

Random Forest classification algorithm was constructed to identify a set of determinants able to separate those who are not willing to vaccinate from those who are. The model started with a list of 85 features and predicted the willingness of subjects in the hold-out dataset with 90.2% balanced accuracy (solid line), which is an average of 90.7% sensitivity (dashed line closed circles) and 89.7% specificity (dashed line open circles). The balanced accuracy remained near constant when testing the recursively reduced models, up to the model with 9 remaining features (i.e. 5 questions with a total of 9 answers, see inserted table). This final model showed an 89.5% balanced accuracy with 87.4% sensitivity and 91.6% specificity. Further reduction, removing the least important feature from the set of 9 (i.e. ‘Neutral’ to ‘Once approved, I believe a COVID-19 vaccine will help protect myself and others’), resulted in a 12.3 percent point reduction in balanced accuracy primarily due to misclassification of the not willing to vaccinate (Specificity = 55.6%).