Observational, cross-sectional, and analytical study with a quantitative method, conducted through an online survey between May 2023 and January 2024.

The northern region of Brazil presents one of the lowest Human Development Indexes (HDI) in the country, ranging from 0.645 to 0.708. States like Amapá and Roraima lead the region with the highest HDI scores, while Amazonas and Pará report lower values. Although the region is rich in natural resources, disparities in access to public services and geographic isolation contribute to its lower human development performance, which reflects inequalities in areas such as health, education, and income (18, 19).

The population is predominantly composed of mixed-race individuals, and away from urban centers, there are groups of riverside communities, quilombola communities, and indigenous peoples who often live in geographic isolation due to the large territorial size of the states and insufficient transportation infrastructure. This affects socioeconomic development and hinders access to basic health services (Figure 1) (18, 19).

The study included individuals aged 18 years or older, residing in the states of Acre, Amapá, Amazonas, Pará, Rondônia, Roraima, and Tocantins who self-reported having a laboratory-confirmed diagnosis of COVID-19 were included, and experienced the persistence of any physical and/or mental health symptoms for 4 weeks or more after COVID-19. The exclusion criteria were individuals without internet access or an electronic device to complete the questionnaire.

The calculations were performed using the EpiInfo version 7.2.5.0 in the StatCalc module.

For the determination of the sample calculation, the following parameters were adopted: the number of recovered COVID-19 cases in the northern region, which totaled 2,827,897 accumulated cases (20–27); a long COVID case frequency of 34.9% (28); a margin of error of 5%; and a confidence interval of 95%, resulting in a minimum sample size of 349 participants.

After the sample calculation, a proportional stratification was carried out for each state in the northern region. The proportion of recovered COVID-19 cases was calculated for each state by dividing the number of recovered cases in that state by the total number of recovered cases in the northern region. Subsequently, the proportion of recovered COVID-19 cases for each state was multiplied by the initial sample calculation (349). The final sample consisted of: 104 (Pará), 20 (Acre), 22 (Amapá), 77 (Amazonas), 58 (Rondônia), 23 (Roraima), and 45 (Tocantins).

For data collection, we used an adapted version of the “Global COVID-19 Clinical Platform: Case Report Form (CRF) for Post COVID condition (Post COVID-19 CRF)” from the World Health Organization (WHO) (29). To adapt this instrument, we held five meetings with experts on the subject to define the information to be collected. Additionally, the research team modified the language used in the questions to make it more accessible to the general population. To enhance the fluency of the questionnaire, we minimized unnecessary reading effort, avoided redundancies, reduced the length of questions, and established a cohesive and coherent order for the questions. We also created a visually comfortable layout suitable for the various devices that participants might use (Supplementary material 1).

On average, participants took 15 min to complete the questionnaire. The questionnaire consisted of nine modules, but for this study, only four were utilized: demographic characterization, information related to SARS-CoV-2 infection, symptoms associated with long COVID, and use of health services. The questions in these domains were grouped according to the three main components of Andersen’s theory: predisposing factors, facilitating resources, and need for care/services.

In the predisposing factors, the following variables were used: Sex; Age; What is your highest level of education? Traditional Population; What is your current occupation? Skin color; Marital Status; Have you heard of the term Long COVID? Have you received any vaccines against COVID-19? How many doses? Did you receive a COVID-19 diagnosis before being vaccinated? How many times have you had a confirmed COVID-19 diagnosis?

In the facilitating resources, the following variables were used: Do you live in an Urban or Rural Area?; Individual income; Are you a beneficiary of Government Social Programs?

In the need for services component, the following variables were included: Before having COVID-19, did you have a diagnosis of any chronic disease?; Symptoms - Pain (General/Muscular/Joints); Symptoms Related to Mental Health (Fatigue/Suicidal Thoughts/Anxiety/Sleep Disturbances/Mood Changes/Depression/Stress/Suicide Attempt); Cardiovascular Symptoms (Heart Attack/Palpitations/Increased Heart Rate/Thrombosis/Arrhythmia); Dermatological Changes (Hair Loss/Dermatitis/Skin Changes); Loss of Taste; Loss of Smell; How would you rate your ability to care for yourself after the pandemic?; Regarding the episode you classified, how was your treatment/follow-up?; Thinking about your most severe COVID-19 infection, how would you classify it?; Did you receive Antibiotics?; Did you receive Antivirals?; Did you receive Ivermectin?; Did you receive Chloroquine?; Did you receive Home Remedies (herbal tea or bottled remedies)?; Respiratory Symptoms (Fatigue/Dyspnea/Shortness of Breath/Cough/Pulmonary Fibrosis/Pulmonary Embolism); Neurological Symptoms (Lack of Attention/Memory Loss-Forgetfulness/Headache/Dizziness); Symptoms of Non-Communicable Chronic Diseases (Hypertension/Diabetes); General Malaise Symptoms (Sweating/Tremors); Were you hospitalized for any of these symptoms?

In this study, we defined traditional populations of the Amazon as those living in the waterways, fields, and forests, including quilombola communities (Afro-descendants residing in specific territories), indigenous peoples, and riverside dwellers, in accordance with the concept established by the National Policy for Comprehensive Health of Rural and Forest Populations (30).

The study team was trained for data collection by experienced specialists in online survey research. Following this, data collectors (students from public and private universities in health-related undergraduate programs) were recruited through social media and college outreach. These individuals were properly trained for data collection through online meetings with the study team (master’s students and undergraduate research assistants). In total, seven universities in the state of Pará were visited, 51 online meetings were held with students from the northern region, and 257 data collectors were trained.

To reach the target population for the research, the Respondent Driven Sampling (RDS) methodology was employed, commonly used to access key populations that are hard to reach or when epidemiological data on a particular disease is not available (31). This methodology begins with a small group of participants, referred to as “seeds,” who contact individuals within their social circles to recruit new participants. These new participants, in turn, recruit others, forming chains or “waves” of recruitment, which allows for a more heterogeneous and representative sample of the population (31).

Thus, by adapting the method for virtual environments, each data collector contacted 10 individuals who had COVID-19 and sent the survey link via WhatsApp. After participants responded, each data collector requested the contacts of other individuals, thereby creating referral chains that extended the recruitment cycles or waves, increasing the number of participants (31, 32). Each data collector recorded in an Excel spreadsheet the number of participants they contacted and how many individuals were referred. This process continued until the researchers achieved the minimum sample size for the study.

Participants received a link via WhatsApp from the data collectors, containing a message written in accessible language with information about the nature and confidentiality of the research. By clicking the link, they were directed to the Research Electronic Data Capture (REDCap®) platform, where they accessed the Informed Consent Form (ICF) and the adapted questionnaire.

The questionnaires were hosted on the REDCap® platform, which is specifically designed for online data management. This platform provides features and tools that enhance organizational efficiency and ensure secure data storage. It is important to highlight that, given the nature of the online survey, all the information collected was self-reported.

The outcome variable of the study is based on the question ‘Did you seek any healthcare service to treat these symptoms?’, with ‘yes’ or ‘no’ as response options. This variable is binary in nature, with the event of interest being the ‘yes’ responses. The independent variables are those present in the questionnaire.

The data collected were extracted from the REDCap® platform and exported to Microsoft Excel®, where descriptive statistics (absolute and relative frequencies) were calculated. Responses defined as no information (NI) were excluded from the analyses, and no percentage was calculated.

The primary hypothesis of the study, ‘there is an association between clinical and demographic aspects and the use of healthcare services among people with long COVID,’ was tested using multiple logistic regression, where the independent variables were associated with the variable ‘Did you seek any healthcare service to investigate and/or treat these symptoms?’ Given the dichotomous nature of the dependent variable, univariate logistic regression model was initially used to assess the association between the independent and dependent variables.

Variables with a p-value <0.20 were selected. The variable ‘Mental Health-Related Symptoms’ had only positive responses, making its inclusion in the regression impossible.

The selected variables were analyzed through multiple logistic regression using the stepwise backward method, which involves using the base model with all previously inserted variables and then removing factors step by step to achieve the lowest Akaike Information Criterion (AIC).

Finally, variables with p > 0.05 were excluded from the final model. To assess the model’s quality, Odds Ratio (OR), Confidence Interval (CI), Akaike Information Criterion (AIC), and Variance Inflation Factor (VIF) were used. The analyses were performed using RStudio software version 2024.04.1 + 748, utilizing the following packages: base; readxl; dplyr; caret; MASS; car; rcompanion, and DescTools.

All requirements set forth by Resolution 466 of 2012 of the National Health Council of Brazil were followed, along with the principles established in the General Personal Data Protection Law No. 13,709 of 2018, particularly regarding personal data processing activities, as detailed in articles 6 and 7. The Declaration of Helsinki was also adhered to. The study received approval from the Research Ethics Committee of the Federal University of Paraíba, Lauro Wanderley University Hospital, under protocol number 5.826.893 and CAAE: 65929522.1.0000.5183.

All participants signed the Free and Informed Consent Form. The authors did not conduct experiments involving humans and/or use human tissue samples or human data. Additionally, there was no direct contact between the authors and the study participants, as the questionnaire was administered online, and the data were securely stored in the REDCap® platform, which ensures privacy and data security.

This study included 364 people with long COVID, of whom only 167 (45.88%) sought healthcare services to treat the symptoms of this clinical condition. The average age of participants was 35.19 years (SD = 13.31), with 48% (173) in the age range of 18 to 32 years, 60% (217) being women, 63% (226) reporting they were not in a stable union/marriage, 70.8% (257) self-identifying as mixed race, Black, or Indigenous, and 89% (325) not being part of any traditional populations (quilombola/Indigenous/riverside) (Table 1).

Table 1 presents the results of univariate logistic regression model showing the association between predisposing factors for accessing healthcare to treat long COVID in northern Brazil. The results show that female participants were twice as likely to access healthcare for long COVID (AOR = 2.16; p = 0.01). Participants already familiar with long COVID were nearly twice as likely to seek healthcare to treat its related symptoms (AOR = 1.97; p = 0.02).

Univariate logistic regression model showed that being aged 18 to 32 years (AOR = 1.69; p = 0.01) and being diagnosed with COVID-19 before vaccination (AOR = 1.64; p = 0.02) were significant factors. However, education level, marital status, current occupation, skin color, belonging to traditional populations, the number of COVID-19 diagnoses, receiving a vaccine, and the number of doses were not associated with accessing healthcare.

Regarding facilitating resources (Table 2), the results show that individuals with higher personal income (AOR = 1.53; p = 0.04) were more likely to access healthcare to treat long COVID. Being a beneficiary of government income distribution programs and residential areas did not influence healthcare access.

In terms of service needs (Table 3), participants with a chronic disease diagnosis prior to COVID-19 were twice as likely to access healthcare for long COVID (AOR = 2.44; p = 0.01). Regarding medication therapy, participants who received antibiotics for treating a more severe COVID-19 infection were three times more likely to access healthcare (AOR = 3.77; p = 0.01). Additionally, participants who received antivirals (AOR = 2.57; p = 0.01), ivermectin (AOR = 2.19; p = 0.01), and chloroquine (AOR = 2.32; p = 0.01) were twice as likely to access healthcare.

Participants who treated their more severe COVID-19 infection under the supervision of a healthcare professional were three times more likely to access healthcare for long COVID (AOR = 3.06; p = 0.01). Furthermore, participants hospitalized due to any long COVID symptoms had eight times the likelihood of accessing healthcare (AOR = 8.67; p = 0.01). A lower odds ratio in participants with loss of smell indicates that they were less likely to access healthcare (AOR = 0.5; p = 0.02).

A multiple logistic regression (stepwise backward) was conducted to identify access components. All variables with a p-value <0.20 were selected and tested together in the model, including predisposing characteristics (‘sex,’ ‘how many vaccine doses?’ ‘How many times have you had a confirmed COVID-19 diagnosis?’), facilitating resources (‘income’), and service needs (‘chronic disease diagnosis before COVID-19,’ ‘classification of the most severe COVID-19 infection,’ ‘how was the follow-up/treatment of the most severe COVID-19 infection?’ ‘Received antibiotics?’ ‘Received antivirals?’ ‘Received ivermectin?’ ‘Received chloroquine?’ ‘Received home remedies [herbal teas or concoctions]?’ ‘Symptoms - pain [general/muscular/joint],’ ‘cardiovascular symptoms,’ ‘dermatological changes,’ ‘general malaise symptoms,’ ‘loss of smell,’ ‘loss of taste,’ ‘hospitalized due to any long COVID symptoms?’).

The final analysis model (Table 4) showed that participants with dermatological changes had twice the likelihood of accessing healthcare to treat long COVID (AOR = 2.57; p = 0.01); participants with a chronic disease diagnosis before COVID-19 were five times more likely to access healthcare for long COVID (AOR = 5.62; p = 0.04); participants who treated their more severe COVID-19 infection with healthcare professional supervision had almost five times the likelihood of accessing healthcare for long COVID (AOR = 4.97; p = 0.01) compared to those who treated it on their own; and participants who used antibiotics during their more severe COVID-19 infection were three times more likely to access healthcare for long COVID (AOR = 3.24; p = 0.01).

The study’s limitations are related to the need for confirmed COVID-19 diagnosis through rapid tests or laboratory exams (limiting participant recruitment) and the inherent limitations of online surveys, such as the homophily and self-selection phenomenon. The first refers to participants with similar sociodemographic conditions, and the second to the lack of control over respondents, which influenced the average age, income, and education level of the study participants. Moreover, the use of electronic questionnaires may have limited the participation of older individuals and those with lower levels of education.

It is crucial to acknowledge that the cross-sectional observational design of this study does not facilitate the establishment of causal relationships between dependent and independent variables. Consequently, the characteristics of the population and the healthcare system examined in this study should be taken into account when generalizing the findings.

The web survey allowed us to understand access to health services among Amazon residents who had symptoms of long COVID, which would be difficult and costly if the questionnaire were applied in person. However, the limitation was having the self-report of the laboratory diagnosis of COVID-19.