Did text-based news-media coverage about the COVID-19 pandemic increase vaccine uptake? A population-based study in Alaska

ABSTRACT

COVID-19 vaccinations protect against severe infection, hospitalisation, and death. News media can be an important source of information for the public during a health crisis. This study explores the extent to which local or statewide text-based news coverage of the pandemic was related to the uptake of initial doses of COVID-19 vaccines among adults in Alaska. Multilevel modelling was employed to explore the association between news media intensity and vaccine uptake rates across boroughs and census areas, while controlling for relevant covariates. Results suggest that the intensity of news media did not significantly influence vaccine uptake during the majority of this time period and had a negative affect during the Delta-surge in the fall of 2021. However, the political lean and median age of boroughs or census areas were significantly associated with vaccine uptake. Race, poverty, or education were not significant determinants of vaccine uptake suggesting there are unique differences in Alaska compared to the U.S., particularly amongst Alaska Native people. The political environment in Alaska surrounding the pandemic was polarized. Future research in communications and channels that can cut through this polarized and politicized environment, and reach younger adults is needed.

KEYWORDS:

• COVID-19
• vaccines
• news media
• Alaska
• politicization
• pandemic
• Alaska native

Introduction

The development and availability of COVID-19 vaccines has meant many COVID-19 related deaths and serious illnesses requiring hospitalisation were now more preventable [1] Vaccines were seen as the primary means of putting an end to the pandemic due to their ability to reduce the morbidity and mortality of COVID-19 for individuals as well as reducing the burden on health care systems [2]. Although vaccine breakthrough infections have and continue to occur, research has demonstrated that those who are vaccinated are far less likely to need hospitalisation than those who are not vaccinated [3].

Vaccine hesitancy

Despite the increasing availability of vaccines across the US, many people have expressed hesitancy to pursue vaccinations to prevent severe outcomes from contracting COVID-19 [4]. Vaccine hesitancy often stems from a multitude of reasons. Black, Indigenous, and people of colour (BIPOC) have historical reasons for distrusting the western medical systems and engaging in research due to previous exploitations [5,6], but importantly, they often experience present-day racism that affect the trust they have in medical expertise [7]. Partisan political affiliations have sharply divided populations in the US on the pandemic, with Republican-leaning counties and states tending to be less in favour of many public health interventions, including vaccinations [8,9]. National data suggest lower rates of vaccine uptake are associated with younger populations and those who report lower levels of education within the context of western educational sytems [10]. In addition to these factors, rapid circulation of misinformation and disinformation online about the COVID-19 vaccines has also stalled vaccine uptake [11].

Emerging outbreaks and news media

Traditional news media coverage has historically been the most commonly used source of information during the emerging outbreaks of infectious diseases [12–14]. Media exposure during emerging outbreaks of novel infectious diseases can have a powerful impact on behaviour [15]. In the case of the current novel coronavirus pandemic, intense media coverage initially was associated with vast reductions in mobility of people through their communities prior to public restrictions on such mobility or business closures [16]. A modelling study, validated with local epidemiological data in Hubei, China, during the early period of the present pandemic, showed that when models included a higher rate of news media information, viral transmission was reduced and thus fewer cases resulted, and the number of cases needing hospitalisation was delayed [17]. However, models in which the news media information rate was reduced resulted in an earlier peak of people needing hospitalisation and an increase in confirmed cases [17].

The H1N1 Influenza pandemic in 2009 also offers important knowledge development around how media contributes to public behaviours, including vaccine uptake [18]. In this study of five European countries, media attention on the pandemic diverged from the epidemiological trends of cases and deaths. News coverage dropped off while case numbers remained high, deaths increased, and vaccine uptake was lower than expected, and in some cases, even lower than in previous years uptake of annual influenza vaccines [18].

Other infectious diseases, vaccines, and the media

In terms of understanding whether media coverage can increase vaccine uptake of existing diseases, influenza vaccines offer a unique opportunity to study the relationship between media and vaccine uptake, because influenza vaccines are currently recommended annually. Increased media coverage was shown to be positively associated with vaccine uptake during the early (October) and later (January and on) periods of the typical influenza season, but not in November and December amongst people 65 years and older [19]. This finding was strongest when the vaccine was noted to be a good match to the virus in circulation [19]. For every 100 news media headlines mentioning influenza or flu, a 0.3% increase in vaccine uptake was noted amongst states in this study. These findings suggest that the severity of the flu season mattered, as well as that less attention is paid to news media coverage during the holiday season [19].

Media coverage can influence parental decision-making regarding childhood vaccinations as well. Coverage in the Danish media about negative and unfounded side-effects from the Human Papilloma Virus (HPV) vaccine was shown to be related to a rapid decline in vaccine uptake among 12-year-old girls in Denmark after an initially successful rollout of the HPV vaccine [20]. Only 54% of 12-year-old girls born in 2003 took up the vaccine, while previously, 90% of 12-year-old girls born between 1998 and 2000 had taken up the vaccine [20]. This is noteworthy as it highlights how negative emotions such as fear can quickly change public sentiment. In another study, a strong correlation between the number of social media posts and news items covering a measles outbreak in a community with low Measles, Mumps, and Rubella (MMR) vaccine uptake was noted [21]. The correlation between social media posts and news items was much stronger than the correlation between the number of social media posts and the actual numbers of measles cases, suggesting that media coverage can influence public sentiment even if the media coverage is not congruent with epidemiological information [21].

COVID-19 vaccine intentions given media exposure

Another recent study in the US investigated whether the main source of news played a role in the intentions of participants to vaccinate against COVID-19, using a nationally representative probability sample of survey panel participants [22]. Of the participants, 83.3% taking in mainstream sources, such as the New York Times or the Washington Post, were likely to vaccinate, while only 58.4% of those utilising conservative news platforms (Fox, Breitbart, etc.), reported they intended to vaccinate [22]. A study based in Germany with 629 adults also showed media, particularly local newspapers, as being both a trusted and an often-used source for health information. Use of those local newspapers as a regular media source was related to increased intentions to vaccinate against COVID-19 [11]. In this study, content experts and health authorities were trusted the most, but were noted to rarely be involved in communication campaigns. A notable factor in this study was that while older people used newspapers and trusted them as a source, younger people tended to report relying on the internet and social network sources more often and this was related to lower intentions to vaccinate. Social networking sources are sites where individuals can network and engage with one another online, such as Facebook or Twitter [23]. These sources are also places where misinformation can be spread rapidly [10,23]. These findings, regarding more traditional news, were replicated in a US non-probability, purposive survey sample of those likely to be in the vaccine priority groups early in the vaccine administration period [24]. Here, when people reported that their main source of news was a traditional media source, such as TV news or newspapers, they were 81% more likely to indicate vaccine acceptance than those not using national newspapers as a main source of information [24]. This work did not adjust for age of participants because no significant relationship between age of participants and vaccine acceptance was found [24].

Vaccine uptake in Alaska

Alaska was the first state to open up COVID-19 vaccine eligibility to all adults on 9 March 2021 [25]. Before this open-eligibility date for vaccines, Alaska often led the US in per capita vaccinations with between 13% and 14% of its population vaccinated in late February and early March 2021 [26,27]. For a period after 9 March until May 2021, Alaska was also a leader within the Arctic, along with Northern Canada, where just under 40% (Alaska) and 50% (Northern Canada) of their respective populations were fully vaccinated in May [28]. The strong collaboration between Alaska’s Tribal Health Care System and the state has been credited with much of this early success [27,29]. Tribes are acknowledged as their own jurisdictions by the federal government and thus were able to receive their own allocation of vaccine doses and decide independently how to distribute and message the vaccine in Tribal jurisdictions [29]. In addition to this self-governance of its healthcare as well as the skill exercised in logistically getting vaccine doses to very remote communities, Alaska Native people and cultural values are unique. Alaska Native people are known to be attuned to prioritising the community, as well as elders and community members valued for their historical knowledge [26,30]. Those elders with historical knowledge are informed by the Alaska Native experience with the 1918 Influenza pandemic during which many rural Alaskan villages suffered high mortality [31]. The cultural loss that resulted from such high mortality has contributed to an urgency around prevention with particular attention towards respiratory viruses, such as COVID-19 [29,31,32].

Despite the initial successes in Alaska’s COVID-19 vaccine rollout, during early summer 2021, the pace of vaccinations in Alaska slowed as did daily COVID-19 case counts. Alaska fell behind other states in the uptake of COVID-19 vaccines and the state’s chief medical officer acknowledged that misinformation and distrust were plaguing vaccination efforts [33].

In July 2021, the Delta variant arrived in Alaska [34] and began circulating, just as Alaskans began to prepare for a return to in-person school in August and after the state’s most populated area Anchorage had ended its emergency declaration [35]. By September of 2021, the Delta-driven surge pushed Alaska’s hospital capacity beyond its limits as the Governor authorised medical facilities to transition into crisis standards of care [36]. As a result, the state offered guidance on how hospitals could ethically deal with limited beds, lack of personnel, and/or depleted resources [36]. September and October of 2021 yielded the highest mortality due to COVID-19 in Alaska. During September and October 2021, 282 deaths were recorded [37], and the rate of deaths rose to 21.3 per 100,000 people in September [38], and still higher in October [39], despite vaccines having been available to all adults for more than 6 months.

Alaska is an important part of the Arctic, as it is equally important to the US as a state positioned within the Arctic and having many natural resources. As a geographically large and isolated state, which is sparsely populated, many of its communities are extremely remote and not connected to urban centres with roads. Within the US, the media environment in Alaska may be substantially different, considering the number of remote communities with limited access to broadband internet or computing devices [40]. Exploring how communications delivered through traditional news media sources, such as print and text-based news, are levied and their impact in the Arctic are an important consideration for public health leaders.

In the US, research has identified common demographic characteristics associated with COVID-19 vaccine hesitancy. Partisan politics are commonly cited; in one study, Republican-leaning US counties were less in favour of vaccinations, and many other COVID-19 pandemic mitigations [8,9]. Lower levels of higher education attainment, as measured in traditional western educational systems [4,10], younger aged adults [10], and people with a lower socioeconomic status [4,41] have also been cited as associated with vaccine hesitancy. Some racial minorities [4,10] have also been noted to be less likely to take up the vaccine. The racial differences during the pandemic in the US have shifted over time though, with many racial minorities overcoming vaccine hesitancy [42,43].

The demographic characteristics of Alaska align with many of these factors, as Alaska leans politically conservative, with 52.6% of the tallied votes in the 2020 presidential election being cast for the Republican candidate [44]. The rate of attainment in higher education is lower in Alaska than the national average, as the 2017–2021 five-year average in the US was 33.7%, while Alaska’s was 30.6% [45]. Historically, Alaska has been demographically young, and although the trend in median age is increasing, Alaska’s 2019 median age was still less than the US median at 35.5 years compared to 38.4 years [46,47]. Alaska is becoming more racially diverse [48] with the largest population centre, the Anchorage municipality known to be home to one of the most racially diverse school districts in the US [49]. Studying the relationship between media and vaccine uptake at the population level, by borough or census area, provides a unique method for investigating the relationship and eliminates some limitations surrounding sampling methods often employed in survey studies.

The present study

The previous research noted here suggest that there may have been an association between news media intensity and vaccine uptake in Alaska that is important to understand. The hypothesis of this present study was that local and statewide text-based news media provided important information to the people of Alaska that could be more relevant than national media and therefore might be related to vaccine uptake. Investigating the relationship between news media coverage of the pandemic and uptake of initial doses of any COVID-19 vaccine in Alaska during the initial months of open eligibility to all adults was the goal of this study.

Materials and methods

This research was deemed non-human subject research by the University of Alaska Anchorage’s Institutional Review Board as the data were publicly available and individuals were not identifiable.

Data sources

For this study, state and national data were compiled from multiple sources for the specific purpose of understanding the relationship between news media intensity and vaccine uptake across all 27 boroughs and census areas in Alaska, while adjusting models for factors that have been identified as associated with vaccine uptake. The State of Alaska’s COVID-19 Information Hub [50] was the source for vaccine uptake and case rates, by borough or census area. The Access World News database [51] was used to collect daily counts of text-based news media coverage. The Alaska Department of Labor & Workforce Development [52] provides population estimates and was used to collate the median age of boroughs and census areas. The Alaska Division of Elections [44] was used to aggregate precinct results from the 2020 Presidential election to the borough and census area level. The Census Bureau’s estimates [53] for race were used to characterise race and diversity within the state. The United States Department of Agriculture’s Economic Research Service [45,54] was used to gather statistics on the attainment of higher education as well as poverty.

Primary outcome variable

The primary outcome variable was the vaccine uptake rate of a first dose of any COVID-19 vaccine among Alaska’s boroughs (county-level equivalent) and census areas (unorganised boroughs). Census areas and boroughs are collectively referred to as boroughs for the remainder of this paper. Data were collected from the state’s COVID-19 Information Hub from 9 March 2021 to 31 October 2021 for Alaskan’s aged 18 and older [50]. The state opened eligibility for the COVID-19 vaccines to all Alaskans over the age of 16 on 9 March making it the ideal date to begin measuring uptake [25].

Primary explanatory variable

News media intensity, as measured through counts of articles pertaining to COVID-19, vaccines, and hospital capacity, was obtained using NewsBank’s Access World News database with filters for the state of Alaska, the time period under study (9 March through 31 October 2021), and having the keywords covid, coronavirus, vaccin*, or hospital* in the headline of the article [51]. Access World News uses a mix of media sources in its database, however the sources available in Alaska were state, and local or regional newspapers and many are available to read online in addition to being printed locally. Sixteen different news sources were identified within this database. News media sources were considered a source for a particular borough or census area if the media source specifically stated it served people from that area or it was also located or headquartered in that geographic region. Daily counts of news media articles were recorded for each borough by combining statewide sources with each borough’s local or regional news source(s) (see Appendix).

Other Covariates

Alaska’s COVID-19 Information Hub was utilised to obtain data on daily case counts to adjust for in models [50]. Seven-day average COVID-19 case rates per 100,000 people were calculated from daily cases reported in each borough based on reported dates for cases. Each day and the previous 6 days were used to calculate the 7-day averages per 100,000 case rates for each borough.

Non-time varying factors, such as race, age, education, poverty, and partisan political affiliation, were operationalised with various data sources reporting this data at the borough level. For race, 2021 data from the Census Bureau estimating the proportion of each borough that self-reported as non-Hispanic white alone was used [53]. Although national data has previously identified slower vaccine uptake among racial minorities [4,10], reporting the non-Hispanic white population proportion allows for an indirect measurement of the proportion of minority races. For age, the estimated median age of each borough was used with data from Alaska’s Department of Labor and Workforce Development [52]. A 5-year average of the proportion of people in each borough completing college, based on American Community Service Survey data, was used to represent education [45]. Income was operationalised by using the proportion of all people in each borough living below the federal poverty designation, reported from the Census Bureau’s Small Area Income and Poverty Estimates (SAIPE scores) [54]. Last, partisan “political lean” for each borough was calculated using the proportion of votes in each borough that went to the Republican candidate in the 2020 presidential election [44]. Alaska does not report its election results at the borough level, therefore, precinct results were used and aggregated to the borough level. Weighting of precinct election results was done using the 2013 state redistricting boundaries and populations, those in place and use at the time this work was completed [55]. Some precincts spilled over into more than one borough, and in these instances, the precinct results were given to the borough containing the largest share of population within the spillover.

Data were organised by borough as reported by the State of Alaska on its COVID-19 Information Hub. Although 29 such regions exist, the state combines some boroughs for its COVID-19 data releases, and consequently, this work followed suit using the 27 boroughs the state does for reporting. For instance, the state combines the Bristol Bay Borough with the Lake and Peninsula Borough for COVID-19 reporting (Table 1); however, most data sources used to collate covariates in this work report each borough independently, so in these cases, data for each borough was weighted based on its 2021 estimated population [52] when combined.

Table 1. Boroughs and census areas in Alaska and shorthand notation used in figures.

Shorthand	Borough or Census Area Name
ABE	Aleutians Borough East
AWC	Aleutians West Census Area
AM	Anchorage Municipality
BBL	Bristol Bay Borough + Lake and Peninsula Borough
BCA	Bethel Census Area
DB	Denali Borough
DCA	Dillingham Census Area
FNB	Fairbanks North Star Borough
HB	Haines Borough
JCB	Juneau City and Borough
KPB	Kenai Peninsula Borough
KGB	Ketchikan Gateway Borough
KIB	Kodiak Island Borough
KCA	Kusilvak Census Area
MSB	Matanuska-Susitna Borough
NCA	Nome Census Area
NSB	North Slope Borough
NAB	Northwest Arctic Borough
PB	Petersburg Borough
PWH	Prince of Wales-Hyder Census Area
SCB	Sitka City and Borough
SM	Skagway Municipality
SFC	Southeast Fairbanks Census Area
VCC	Copper River Census Area + Chugach Census Area
WCB	Wrangell City and Borough
YKC	Yukon-Koyukuk Census Area
YHA	Yakutat City and Borough + Hoonah-Angoon Census Area

The Valdez-Cordova Census Area (VCC) was formerly its own census area but is now broken into the Copper River Census Area and the Chugach Census Area.

Data analysis

All analyses were performed using R statistical software [56]. Models were constructed using the lme4 package, data cleaning using the dplyr and zoo packages, while graphics were done with the ggplot2 package [57–60].

First, monthly news media intensity was explored and summary statistics reported (Table 2). For the analytical modelling, a linear multilevel modelling approach was used in this population-based study. Non-independence and autocorrelation noted in vaccine uptake rates and covariation at the borough level suggested that this approach was appropriate for modelling in order to answer the research question about how news media intensity may be related to vaccine uptake [61]. For modelling, news media counts were lagged by 3 days behind daily case rates and vaccine uptake was then lagged by 9 days from daily case rates. This was done to account for what was hypothesised as a reasonable amount of time for changing case rates to be reflected in news media reports as well as for both case rates and news media reports to potentially impact vaccine uptake.

Table 2. Descriptive findings from the news-media intensity by month.

Month	News Article Totals
March  Mean  Median  Min  Max  Total	11.5 13 2 20 265
April  Mean  Median  Min  Max  Total	11.5 12 1 24 345
May  Mean  Median  Min  Max  Total	8.3 8 1 17 258
June  Mean  Median  Min  Max  Total	5.1 5.5 1 13 153
July  Mean  Median  Min  Max  Total	8.3 7 1 21 257
August  Mean  Median  Min  Max  Total	13.9 15 3 30 431
September  Mean  Median  Min  Max  Total	16.1 17 3 34 484
October  Mean  Median  Min  Max  Total	12.3 13 1 26 381

The first step in model construction was setting the random effects structure, utilising Reduced Estimation Maximum Likelihood (REML) estimation. Three models with three different random effect structures were developed. The first utilised both the random effects representing spatial covariation (borough variable) and the other representing temporal covariation (time in days), another model held just the control of spatial covariation and the last with just the temporal covariation.

After selecting the random effect structure utilising Akaike Information Criterion (AICs), the fixed effect structure was constructed using maximum likelihood estimates. Nested models were developed by increasing the number of fixed effect variables beyond the key explanatory variable, news media intensity. Each nested pair of models were then compared with a likelihood ratio test (X² statistic) and when the added complexity of each variable offered more inference into understanding what was related to the rise in vaccine uptake, as indicated by a p-value less than or equal to 0.05, the variable was added to the model. Tables 3, 4, and 5 detail this process.

Table 3. Overall time period model construction and selection process.

Random Effects	Fixed Effects	Log-likelihood (LL)		AIC		ΔAIC
Random Effects Structure (REML estimation)
B + D	All	14282		−28542		0
D	All	5791		−11562		16980
B	All	8376		−16732		11810
Fixed Effects Structure (Maximum Likelihood estimation)
Random Effects	Fixed Effects	LL	AIC	ΔAIC	X^2	df	p (LRT)
B + D B + D	N N + CR	14317 14317	−28624 −28622	17.23 19.15	0.1	5 6	0.7756
B + D B + D	N N + PL	14317 14324	−28624 −28635	5.90	13.3	5 6	0.0003
B + D B + D	N + PL N + PL + A	14324 14328	−28635 −28641	0	7.9	6 7	0.0049
B + D B + D	N + PL + A N + PL + A + E	14328 14328	−28641 −28639	1.71	0.3	7 8	0.5881
B + D B + D	N + PL + A N + PL + A + P	14328 14328	−28641 −28639	1.81	0.2	7 8	0.6616
B + D B + D	N + PL + A N + PL + A + R	14328 14328	−28641 −28639	1.89	0.1	7 8	0.7383

Note for Tables 3, 4 , and 5: Random effects are represented by B (borough or census area) and D (time in days). Fixed effects are represented by N (news media intensity), CR (case rate), PL (political lean), A (age), P (poverty), R (race), and E (education). For each nested set of models adding in fixed effects, the log-likelihood of that model (LL), the AIC, AIC differences (ΔAIC), the likelihood ratio test statistic (X²), associated degrees of freedom (df) with each model, and the P-value from the likelihood ratio test are provided. A model is advanced to the next row of nested models, if the added complexity of that model added significant inference to understanding vaccine uptake (the likelihood ratio test statistic was significant with a P-value less than or equal to 0.05). The best-supported model is indicated in bold, both for the initial setting of random effects and for the overall model.

Table 4. Delta-surge time period model construction and selection process.

Random Effects	Fixed Effects	LL		AIC		ΔAIC
Random Effects Structure (REML estimation)
B + D	All	7805		−15589		0
D	All	1981		−3943		11646
B	All	5794		−11568		4021
Fixed Effects Structure (Maximum Likelihood estimation)
Random Effects	Fixed Effects	LL	AIC	ΔAIC	X^2	df	p (LRT)
B + D B + D	N N + CR	7806 7844	−15603 −15676	88.24 15.15	75.1	5 6	<2.2e-16
B + D B + D	N + CR N + CR + PL	7844 7850	−15676 −15686	4.84	12.3	6 7	0.0004
B + D B + D	N + CR + PL N + CR + PL + A	7850 7854	−15686 −15691	0`	6.9	7 8	0.0088
B + D B + D	N + CR + PL + A N + CR + PL + A + E	7854 7854	−15691 −15689	1.82	0.2	8 9	0.6581
B + D B + D	N + CR + PL + A N + CR + PL + A + P	7854 7854	−15691 −15689	1.89	0.1	8 9	0.7235
B + D B + D	N + CR + PL + A N + CR + PL + A + R	7854 7854	−15691 −15689	1.97	0.0	8 9	0.8351

Table 5. First 38 days model construction and selection process.

Random Effects	Fixed Effects	LL		AIC		ΔAIC
Random Effects Structure (REML estimation)
B + D	All	1481		−2939		0
D	All	804		−1588		1351
B	All	1192		−2364		575
Fixed Effects Structure (Maximum Likelihood estimation)
Random Effects	Fixed Effects	LL	AIC	ΔAIC	X^2	df	p (LRT)
B + D B + D	N N + CR	1507 1509	−3005 −3007	24.06 22.06	4.0	5 6	0.0446
B + D B + D	N + CR N + CR + PL	1509 1518	−3007 −3021	7.83	16.3	6 7	0.0000
B + D B + D	N + CR + PL N + CR + PL + A	1518 1523	−3021 −3029	0	9.9	7 8	0.0017
B + D B + D	N + CR + PL + A N + CR + PL + A + E	1523 1523	−3029 −3028	1.22	0.8	8 9	0.3621
B + D B + D	N + CR + PL + A N + CR + PL + A + P	1523 1523	−3029 −3028	1.56	0.5	8 9	0.4839
B + D B + D	N + CR + PL + A N + CR + PL + A + R	1523 1523	−3029 −3027	1.98	0.1	8 9	0.802

Additional modelling

Within the time period studied, two smaller time periods were distinctly different from the overall time period. First, as can be seen in Figure 1, the rates of vaccine uptake in boroughs rose much more steeply during the first 38 days of this time period. Prior to this period, COVID-19 vaccines were only available to adults aged 55 and older, health care and essential workers, educators, and adults over the age of 16 with a health condition that put them at higher risk [25]. The steeper rise in vaccine uptake during this time period, when compared to the slower rise later in the time period under study, led to questions about how news may have been different during this period, or whether differing demographic characteristics would define vaccine uptake. The data was subset to include the first 38 days of this time period and modelled exactly the same as the overall time period. During the Delta surge, case rates rose steeply in Alaska (see Figure 2) and eventually led to increased hospitalisations and deaths. In the interest of understanding whether different epidemiological situations resulted in different factors explaining vaccine uptake, another model was developed to represent this time period. The data was subset when the statewide 7-day case rate rose to 50 cases per 100,000 people and was modelled in the same fashion as previously outlined for the entire time frame. Fifty cases per 100,000 was categorised as “substantial” transmission at that point in time [62]. This time period included the days from 17 August to 31 October. This model is referred to as the Delta-surge model.

Figure 1. Uptake of an initial dose of any COVID-19 vaccine in Alaska by borough or census area from March 9^th – October 31^st, 2021. the first vertical black line represents where data was subset in order to model the first 38 days. The second vertical black line represents where the data was subset to model vaccine uptake during the Delta variant-driven surge.

Figure 2. Seven-day average COVID-19 case rates per 100,000 people by borough and census area in Alaska from March 9^th – October 31^st, 2021. the first vertical black line represents where data was subset in order to model the first 38 days. The second vertical black line represents where the data was subset to model vaccine uptake during the Delta variant-driven surge.

Model assessment

The models were assessed and interpreted statistically as well as practically. As expected, vaccine uptake over time in Alaska’s boroughs is an autocorrelated series of values. While the models addressed this through the use of random effects to account for the lack of independence in vaccine uptake from one day to the next as well as covariation between boroughs, and lags were applied to appropriate variables, autocorrelation remains in the model residuals.

As Figure 1 highlights, vaccine uptake through the entire time period was not linear. After modelling the data with and without the first 28 days of data, the very similar results suggest that the models were robust to the violation of this linear assumption.

Results

Descriptive results

Table 2 provides information about how news media intensity changed throughout the study period. September 2021 had the most total articles with 484, while June had the fewest. News media intensity eased from March to April and hit a low median of 5.5 articles per day in June. In July, a 27% increase in the median number of articles daily was noted with 7 articles per day. A 114% increase in news media intensity was noted in the median articles per day from July to August with an increase from 7 to 15 articles daily. Finally, a 13% increase was noted in September, up to a 17 article per day median, and then in October, intensity began decreasing with a 24% drop down to a 13 article per day median.

Table 6 highlights the summary of non-time varying variables used to characterise the differing demographics of boroughs in Alaska. The range of values representing racial diversity in Alaska, as measured by the proportion of people self-identifying as non-Hispanic white alone is between 4% in the Kusilvak Census Area and up to 84% in the Skagway Municipality. The range of the percentage of adults earning a college degree in boroughs was between 4% in the Kusilvak Census Area to 39% in the Denali Borough as well as in the City and Borough of Juneau. For income, reported by the percentage of people in each borough falling below the federal poverty level, estimates ranged from a low of 4% in the Skagway Municipality to a high of 28% in the Kusilvak Census Area. The political lean of boroughs, reported as the percentage of the votes in 2020 going to the Republican candidate for President, had a range between a low of 29% of votes in the Kusilvak Census Area, and a high of 82% was noted in the Southeast Fairbanks Census Area. Last, the range in the median age of boroughs is spread between a low of 24 years in the Kusilvak Census Area and a high of 47 years in the Haines Borough, the Wrangell City and Borough, as well as the combination of the Yakutat City and Borough with the Hoonah-Angoon Census Area.

Table 6. Non-time varying borough-level demographic variables.

Borough or Census Area	Represents the Percentage of the Population				Median Age
	Race Non-Hispanic White	College Degree Attainment	Living in Poverty	2020 Votes for Republican Presidential Candidate
ABE	22	17	15	57	45
AWC	32	17	8	47	40
AM	63	37	8	60	35
BBL	31	24	15	49	37
BCA	9	11	25	35	28
DB	81	39	6	62	43
DCA	15	19	19	46	31
FNB	76	32	7	66	35
HB	80	28	10	54	47
JCB	68	39	8	52	39
KPB	83	26	11	75	42
KGB	66	26	8	65	40
KIB	53	30	8	65	36
KCA	4	3	28	29	24
MSB	82	22	9	80	37
NCA	15	17	19	36	30
NSB	30	15	11	47	34
NAB	11	15	19	42	30
PB	74	26	7	72	43
PWH	45	17	16	54	42
SCB	66	34	7	55	41
SM	84	24	4	37	42
SFC	77	21	12	82	38
VCC	71	30	9	66	39
WCB	66	17	11	72	47
YKC	23	13	21	34	37
YHA	47	19	13	44	47

Analytic results

News media intensity during the first 38 days was not statistically related to the increase in vaccine uptake. However, a small, negative affect was noted during the Delta surge model. Finally, news media intensity through the overall time period of the study was not significantly related to the increase in vaccine uptake. The complete results of point estimates for variables identified in our model selection process and 95% confidence intervals are shown in Table 7. In the overall time-period findings, the effect size for every increase in the number of daily news media articles, there was a −0.0001% decrease in vaccine uptake daily, very near to zero and the 95% confidence intervals included zero, indicating that the effect size may be due to random chance. In the first 38 days, the effect size was also negative, at −0.13%, and the confidence intervals, continued to include zero. During the Delta-driven surge timeframe, news media intensity yielded a negative effect size of −0.02%; however, the 95% confidence intervals were both negative, at (−0.045%, −0.004%).

Table 7. First 38 day, overall, and delta-surge model coefficient findings and 95% confidence intervals.

Variable	Effect Size	95% Confidence Interval
March 9, 2021 – April 15, 2021 (First 38 Days Findings)
News-media Intensity	−0.13%	(−0.331%, 0.069%)
Political Lean	−0.76%	(−1.005%, −0.525%)
Median Age	0.96%	(0.382%, 1.532%)
Case Rate	0.02%	(0.002%, 0.040%)
March 9, 2021 – October 31, 2021 (Overall Findings)
News-media Intensity	−0.0001%	(−0.068%, 0.068%)
Political Lean	−0.74%	(−1.01%, −0.47%)
Median Age	0.95%	(0.30%, 1.60%)
August 17, 2021 – October 31, 2021 (Delta Surge Findings)
News-media Intensity	−0.02%	(−0.045%, −0.004%)
Political Lean	−0.73%	(−0.995%, −0.455%)
Median Age	0.92%	(0.276%, 1.573%)
Case Rate	0.002%	(0.0014%, 0.0022%)

The best fitting models for the all three time periods included random multilevel intercept effects controlling for borough as well as time in days. All three models also resulted in fixed effects of median age, and the political lean of boroughs. For every increase of one year in median age of a borough, there was nearly an 1% increase in vaccine uptake during all three time periods. Additionally, for every 1% increase in a borough’s votes going to the Republican candidate in the 2020 presidential election, there was close to a 0.75% decrease in vaccine uptake during all three time periods. Increasing COVID-19 case rates in boroughs led to increasing vaccine uptake during the first 38 days as well as during the Delta-surge period. However, during the overall period, the effect of case rates washed out and were not selected into the final model. The effect of increasing case rates was strongest during the first 38 days, where the model estimated that if a case rate rose to 50 cases per 100,000 people, vaccine uptake would increase by 1% and likewise, if a case rate rose to 100 per 100,000, 2% points in vaccine uptake would follow.

Despite previous work identifying demographic characteristics such as education levels, income and poverty, as well as race explaining vaccine uptake, this study yielded differing results [4,10]. These factors were not selected into final models for any time period. The median age was an important determinant of vaccine uptake though with the largest effect size, in all three time periods. For every increase in median years of age, a nearly 1-point percentage increase was noted in vaccine uptake.

Discussion

With the exception of the Delta-surge period, the intensity of text-based news articles pertaining to the COVID-19 pandemic was not related to vaccine uptake in Alaska during the time period studied. The findings of no effect are contrary to previous findings that have suggested these text-based news articles can influence vaccine uptake as well as other behaviours related to infectious diseases [15,16,19,20]. The negative effect size during the Delta surge suggests that analysing the content of the news coverage may be useful in discerning why the effect of increasing news would have a negative impact on vaccine uptake. One hypothesis for this finding is that unvaccinated people at this point in the pandemic were increasingly more hesitant and the news articles sometimes highlighted the social polarisation around the pandemic. One such article details Alaska’s Governor activating crisis standards of care in light of dwindling hospital resources and strained capacity [36]. The article highlights Facebook comments coming in during the live-streamed press briefing that was viewed by 1350 people, suggesting the data the Governor cited were untrue [36]. Vaccine hesitant people may have dug in their resistance and publicity around the social discord may have made these reactions seem increasingly normalised [63]. Findings during the overall time period and the first 38 days in which news media did not have an effect on vaccine uptake offer different ideas for further investigation into the text-based news media’s impact on vaccine uptake.

First, the political lean of boroughs was related to vaccine uptake with increasing proportions of votes going to the Republican presidential candidate in 2020 related to lower vaccine uptake. In a recent study, print and text-based news about COVID-19 tended to be more politicised and polarised than network or TV-based news about COVID-19 [64]. This may reinforce the findings here, which suggest that a polarised and politicised print and text-based media environment could lead to divergent trends based on the political lean of boroughs in the state. However, this work did not assess the degree of politicisation or polarisation within the text of news articles used in this study. To add complexity to the findings, further research on media consumption habits and COVID-19 behaviours in the US reports that more consumption of news, regardless of the political ideology of individuals or of their media sources lead to more adherent public health behaviours [65]. Readership or actual consumption of news media articles used in this study could not be measured either. However, the lack of a relationship between news media intensity and vaccine uptake in the current study could reflect these findings. Lower consumption of statewide or local text-based news may have been related to political ideology and perhaps a lack of trust in and/or use of these news sources in making decisions, offering another avenue of understanding lower vaccine uptake in the state.

Recent Gallup polls suggest American’s trust in media is near its all-time low with those who report no trust of the media up to 38% and those who have a great deal of trust in media down to 34% [66]. Additionally, partisan political affiliations define how results diverge in this poll with only 14% of Republicans reporting a great deal of trust in the media, while Democrats had much higher trust in the media, with 70% trusting the media a great deal, and 27% of Independents reporting this high level of trust in media [66]. Just as Alaska’s Chief Medical Officer, Dr Zink described the issues of trust and misinformation that plagued Alaska’s vaccination campaign, other national narratives painted a similar picture of problems with trust and vaccine hesitancy in the US [33,42]. New research has identified that measures of interpersonal trust at the state level were associated with COVID-19 outcomes, with lower interpersonal trust equating to worse outcomes, including deaths [9]. While this study cannot measure trust in news media, it may be an important component of further research in Alaska or the Arctic about how public health communication experts can establish trust in their messages regardless of the media channel they utilise.

Next, public health communicators need to reach younger Alaskan adults with compelling reasons to take up COVID-19 vaccinations. A German-based survey study has suggested that text-based news were trusted and utilised among older participants, but they were not the most used or relied upon source for COVID-19 news amongst younger survey participants [11]. In addition, the Chen et al. study of print-based media’s positive impact on influenza vaccine uptake also targeted persons who were 65 years old or older. Taken in conjunction with the present study’s findings, the rationale for vaccinating was likely distinctly different for younger adults and their media habits are likely different than those of older adults. Research with younger adults in Alaska could help identify the concerns that this age group had around COVID-19 to help craft meaningful messaging and communication channels to utilise. While the direct threat to health and the lives of older people was likely a motivating factor for vaccine uptake, younger Alaskans may have had different concerns. Those concerns could have had more to do with closures of schools, loss of jobs, or isolation from typical social activities. As a state with a median age that is less than the median age of the US [52], communication experts in Alaska will need to continue to develop effective techniques with this age group.

The significance of COVID-19 case-rates during the first 38 days of open vaccine eligibility and during the Delta surge in explaining some of the variance in vaccine uptake suggests two different things. Figure 2 details how case rates were lower over the summer. Seasonal habits, such as summer recreation and lower case rates, might have meant people were less aware of case rates during this period. However, during the Delta surge, unvaccinated people may have shifted their perception of risk as hospital capacity broke down and deaths began to climb [36,62], with some taking up the vaccine during this time. Ideally, communication campaigns could become effective at successfully influencing vaccine uptake prior to health crises such as the Delta surge in Alaska though. Research has shown that physicians and other trusted health care professionals are the best source of information during health crises [67–69], but are often not a part of communication campaigns [11,70]. Physicians can also help deliver information that is void of polarised content and political contexts. However, their training in communications may not include addressing highly politicised topics, and they are also inherently people with political opinions and beliefs as well [71].

The successful rollout of COVID-19 vaccines in coordination with Tribal Health Care organisations [28,30] is one such example where vaccines were effectively messaged to Alaska Native people and taken up at a higher rate. A detailed case study of this effort may be useful for other cultural groups or organisational structures to use as a framework for successfully rolling out new vaccines.

Limitations

A major limitation to this study is the reliance on quantitative methods only. A content analysis of news media coverage in Alaska may have yielded greater depth in understanding how text-based news media coverage was or was not related to vaccine uptake in Alaska, but is nonetheless, beyond the scope of this work. Additionally, the source for news media artefacts, Access World News’s database in Alaska is a “proxy” for news media statewide, but not an exact measurement of such, as some print and text-based news-media sources are not included in their database. Last, because of the remaining autocorrelation in the residuals of the models, caution should be taken about interpreting the parameter estimates. While the results practically are justifiable in that they are largely in line with findings about the factors related to COVID-19 vaccine uptake in other regions, the remaining serial correlation may result in some pseudo-replication or inflation of the estimate sizes.

Conclusion

The conclusions here are mixed. There is a lack of solid inference about the relationship between text-based news media intensity and vaccine uptake which warrant further study to understand more clearly. Understanding how media information and environments influence the people of Alaska and the Arctic will be important for future public health issues impacting Arctic nations. Indicators of lower socioeconomic status such as educational levels, income, and race that have been previously identified [4,10] as associated with lower vaccine uptake in the US were not associated with vaccine uptake in Alaska. A likely explanation for the lack of these findings may be because Alaska Native people make up the largest group of minoritized people in Alaska [72]. And while Alaska Native people have been historically marginalised and subsequent indicators of poverty are high [73], they have had amongst the best vaccine uptake within the state, along with other Indigenous peoples of the US [28,30]. While Alaska was different than the US in regard to these demographic measures of socioeconomic status, the findings regarding partisan political lean and age of boroughs were related to vaccine uptake are in line with previous findings in the US [8–10]. The political environment in Alaska surrounding the COVID-19 pandemic was polarised and trust may be a key component for future research in developing communications and channels that can cut through this polarised and politicised environment, whether through traditional, text-based news media sources or other communication channels. Last, work specifically with younger adults in Alaska and the Arctic more broadly in understanding media habits and effective messaging campaigns can guide future communication campaigns with younger adults.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This project was supported financially by two sources:1) This project was supported by the Health Resources and Services Administration (HRSA) of the US. Department of Health and Human Services (HHS) under grant number UB6HP31690 titled Public Health Training Centers. This information or content and conclusions are those of the author and should not be construed as the official position or policy of, nor should any endorsements be inferred by HRSA, HHS or the US. Government.2) Research reported in this publication was supported by an Institutional Development Award (IdeA) from the National Institute of General Medical Sciences of the National Institutes of Health under grant number P20GM103395. The content is solely the responsibility of the authors and does not necessarily reflect the official views of the NIH.

Appendix

Access World News database sources, regional reach and article totals by source.

Table

News-media Source	Regional Reach	Articles Included in Study
Alaska’s News Source	Statewide	603
Anchorage Daily News	Statewide	555
Alaska Journal of Commerce	Statewide	8
Anchorage Press	AM	143
Arctic Sounder	NAB, NSB	52
AP State Wire	Statewide	55
Bristol Bay Times	BBL, DCA	53
Delta Discovery	BCA, KCA	27
Fairbanks Daily News-Miner	DB, FNB, SFC, YKC	285
Juneau Empire	JCB	181
Kodiak Mirror	KIB	71
Mat-Su Frontiersman	MSB	61
Nome Nugget	NCA	44
Northern Light	AM	3
Peninsula Clarion	KPB	188
Sitka Sentinel	SCB	245