Ethics statement

This study was approved by the Institutional Review Board (IRB) of the United States Naval Medical Research Center Unit No. 6 (NAMRU-6) (Protocol #: NAMRU6.2014.0028) in compliance with all applicable federal regulations governing the protection of human subjects. IRB relying agreements were established between NAMRU-6 and Emory University, Tulane University, University of California Davis, University of Rhode Island, San Diego State University, and University of Notre Dame. In addition to IRB approval, investigators obtained host country approval from the Loreto Regional Health Department, which oversees health research in Iquitos. Adult study participants provided written informed consent and a parent or guardian provided informed consent on behalf of child study participants. Children 8–17 years of age provided written assent. The proposed interview schedule was discussed with each participant during the consent process. Study staff always accommodated participant schedules, made every effort to minimize the time needed for the interviews and clinical evaluation, and it was strongly emphasized that participation was voluntary and could be declined at any time. In practice, participants were unavailable on some occasions, but our research group has had a long term relationship with the community in Iquitos and visits by staff during illness are usually valued heavily by our participants. If our physician had any concerns about the severity of illness, they would help facilitate admission into a local hospital, and would continue visits only if requested by participants, which was often the case.

Study area

This study was performed in the Amazon basin city of Iquitos, Peru. Iquitos is a geographically isolated, tropical urban environment with approximately 430,000 inhabitants located near the confluence of the Amazon, Nanay, and Itaya Rivers [37]. The city’s economic structure is highly informal and dynamic, with one-third of economically active individuals either unemployed or informally employed [38]. Iquitos has a fairly homogeneous mestizo culture, with the majority of residents being categorized in socio-economic levels C and D, comparable to middle- and lower middle-class.

Iquitos has been the home of extensive, long-term arboviral research led by the University of California, Davis and U.S. Naval Medical Research Unit No. 6 since 1998 [11,12,39–43]. Human mobility studies paired with detailed epidemiological data make Iquitos an informative site for understanding the dynamics of arbovirus transmission. All four serotypes of DENV have been introduced in Iquitos; however, at any particular time virus transmission is usually dominated by a single serotype [42,44]. Mobility is highly irregular and temporally unstructured, rarely centering around a single location, such as a workplace [40]. Previous research [40] demonstrated that the majority of individual’s movement (~80%) occurs within 1 km of their home.

Study design

Iquitos residents with laboratory-confirmed DENV infection (by polymerase chain reaction [PCR]) were identified and recruited through clinic- and community-based longitudinal febrile surveillance as previously described [12]. At the time of initial diagnostic blood draw, a retrospective semi-structured movement survey (RMS) was verbally administered by trained research assistants (the ‘Movement Team’) to identify locations an individual had visited in the 15 days prior to diagnosis (the exposure period), as well as the visitors they received at their home in the previous three days. For the next seven days a modified survey that focused on activities and personal contacts during the prior 24 hour period (Daily RMS [DRMS]) was carried out [14]. During that seven-day follow-up period, DENV positive (DENV+) individuals were also administered two Quality of Well-Being (QWB) surveys [45], 2–3 days and seven days after the initial PCR-positive blood test result. The QWB survey is a validated instrument used to measure an individual’s health related quality of life during chronic illness [45,46] that uses a weighted algorithm to produce a well-being score between 0 (death) and 1 (optimal health) [45]. On the seventh day after the initial diagnostic blood draw an expenses survey was administered, which focused on the costs incurred during dengue illness. This survey had a subset of questions focusing on the general caregiving behaviors received by a symptomatic participant during his/her illness.

During a follow-up visit, scheduled 30 days after the initial PCR-positive blood test, participants were given “post-illness” RMS and QWB surveys in an effort to measure a baseline for their mobility behavior and well-being in the absence of illness. A second expenses survey was also administered in order to identify costs accrued after the initial day-7 survey. S1 Table provides a description of each survey, including when it was administered, the number of respondents, and a list of the questions that contributed to the current analysis.

Data processing

For each study participant a standardized “day after symptom onset” variable was calculated, rather than referring to daily survey values as occurring on a certain number of days after study enrollment. For example, if a participant provided a blood sample on their third day of symptoms, daily surveys would capture data for days 4–11 after symptom onset. Because DENV cases were captured 1–5 days after onset of their symptoms, daily surveys captured a range of 1–14 days after symptom onset. We focused our analysis on days 1–10 after symptom onset, as only a few individuals (12/71) had data for days 11–14.

For each day after symptom onset, DRMS data were utilized to record (1) how many visitors an individual received at their home, (2) the visitor’s relation to the ill participant, (3) the reason for the visit, (4) if the visitor knew the person was sick at the time of visit, and (5) whether the visitor typically visited the participant at least once a week in the absence of illness (‘routineness’ of the visitor). All data were reported by the subject or his/her parent or guardian. Pre- and post-illness RMS provided a baseline for the number of visitors an individual received.

The ‘Expenses’ survey was given twice, with both datapoints focused on the entire illness period. The datapoints were crosschecked to ensure all caregiving instances were accounted for. Information was provided on (1) whether an individual had someone help care for them (i.e., act as a caregiver), (2) how many people helped, (3) the relation of each helper to the participant, (4) what they helped with, and (5) whether the helper had to take time off from work (i.e., was their work affected). Similarly, the QWB survey was given twice during the symptomatic period. The two datapoints were combined to determine whether individuals felt at any point during their illness that they needed help with daily activities or personal care. An individual’s QWB score, calculated with the weighted algorithm, was also provided for these two time points. An individual’s minimum QWB score across the two time points was utilized as a metric of overall health-related quality of life during illness. Minimum QWB score was considered as a continuous variable as well as a factor variable, where individuals were split into 2 (low/high) or 3 (low/medium/high) groups of equal size based on their minimum QWB score during illness.

Data analysis

Our main goal of data analysis was to examine behavioral changes among social contacts in response to symptomatic dengue illness. Data on visitors and caregivers provided information on a symptomatic individual’s social contacts from inside their house and outside their house.

Details about visits received during the illness period were analyzed using the following variables: (1) presence of visitors (yes/no), (2) number of unique visitors (1/2+), (3) relation of visitors (family/friend/other), (4) number of unique visits by each visitor (1/2+), (5) whether they were ‘routine visitors’ (yes/no), (6) whether they knew that the person they came to visit was ill (yes/no), and (7) the reason for each visit (emotional support/logistic support/other disease reason/unrelated to disease). The first two variables were analyzed at the participant level, whereas (3)-(7) were analyzed at the visitor level. For each variable of interest, descriptive statistics were calculated and comparisons were made based on the gender (male/female) and age (adult:> = 18/child:<18) of the symptomatic individual, the gender/age combination (male adult/male child/female adult/female child), and the health-related quality of life (low/high minimum QWB score during illness). The reason for visiting was also compared based on whether the visitor was ‘routine’ or not. All these comparisons were conducted using Fisher’s Exact test. For non-dichotomous factor variables, (3) and (7), if there was a significant association with gender/age/quality of life during illness/’routineness’, the proportions for each outcome level were compared separately to determine if a certain outcome level was driving the association. Comparisons were also made between presence of visitors pre-illness and during illness to determine the prevalence of losing, gaining, or keeping visitors when ill.

Comparisons were conducted to determine whether the infected person’s gender, age, or quality of life during illness was associated with caregiving behaviors. Details about caregiving were captured for the overall illness period using the following variables: (1) presence of caregivers (yes/no), (2) number of caregivers (1/2), (3) whether the caregiver was a housemate of the sick individual (yes/no), (4) days help was provided, (5) whether the caregiver helped take care of the sick individual (yes/no), (6) whether the caregiver helped around the house (yes/no), (7) whether the caregiver helped by providing money or items (yes/no), and (8) whether their work was negatively affected due to providing help to the sick person (yes/no). Variables (1) and (2) were examined at the participant level and variables (3)-(8) were analyzed at the helper level.

For variables related to caregiving, the association with possible predictor variables was determined using Generalized Linear Models (GLMs). Best-fit models were determined for the logistic response variables of: presence of caregiving (yes/no), number of caregivers (one versus two), relationship of caregiver (housemate or not), if they helped around the house (yes/no), if they helped with buying things or giving money (yes/no), and if the caregiver’s work was affected (yes/no). For each response variable, individual predictor variables considered were the age (child vs. adult) and gender (male vs. female) of the sick individual (as well as an interaction variable for age and gender), whether the sick individual had a high or low number of housemates (split into a binary variable around the median number of housemates), whether they needed help with personal care activities or daily activities at any point during illness (QWB), and minimum QWB score (as a continuous value or using 2/3 groups). If two bivariate regressions had significantly good fits, a model with both variables was considered. Best-fit models were determined using the corrected AIC (AICc), relative likelihood of the model (weight), and a Chi-squared test comparing reduction in residual deviance.

The response variable of whether or not a symptomatic individual received visitors was examined for both the entire illness period and each day of illness. When the entire illness period was considered, GLMs were examined with the same predictor variables as above. When the response variable was the presence of visitors on any given day of illness, Generalized Linear Mixed Models (GLMMs) were used, with the participant ID as a random effect to account for repeated observations [47]. Specific day of illness was considered as a possible predictor variable to determine whether visitor presence changes throughout illness. Variables with set values that did not change during illness (i.e., gender, age, number of housemates, whether help was needed to complete daily or personal care activities, minimum QWB score) were also considered as possible predictors. All statistical analyses were performed in R 3.3.0 statistical computing software [47,48]. Note that multiple correspondence analysis was used to create an SES (socioeconomic) score based on an individual’s employment, education, and home materials (i.e., roof, floor, interior/exterior walls), as well as the presence of household items/services such as computers, “mototaxis” (rickshaws), cable, and internet. This score was stratified into three groups to compare outcomes; however, there were no significant differences found between groups and SES score was not a significant predictor variable in any of the GLM(M)s. This is likely because, outside of the more touristic areas of the city where one might find a higher SES, the SES status of Iquitos’ residents, specifically those in our study areas, is fairly homogeneous: middle to lower-middle class. Based on the possibly small margin between groups and the lack of significance on any outcome variables, we chose not to include the SES score in the results.

Visiting behavior

Approximately one-third (32%) of participants received visitors at some point during their illness. Females had visitors significantly more often than males (47% versus 19%; p = 0.02, Table 1). Of those participants who received visitors, most (48%) had one unique visitor and 70% of visitors came only once during the illness period. Those in the ‘low’ QWB group (lower health-related quality of life) were significantly more likely to receive multiple visitors compared to their ‘high’ QWB counterparts (82% versus 18%; p = 0.009; Table 1). Further, the QWB score was significantly associated with the relationship of the visitor (Fisher’s Exact test, p<0.001; Table 1). Seventy-seven percent of visitors received by those with ‘low’ QWB scores were family members, whereas visitors of those with ‘high’ QWB scores were more likely to be friends (62%) or be in the ‘other’ category (29%). Across all relationships, the majority of visitors were categorized by the ill participant as ‘routine visitors’ (visited at least once a week pre-illness) (87%), who knew that the participant was ill (87%), and were visiting for reasons related to the illness (80%) (Table 2). Additional details on the description of the reasons for visiting are presented in the supplemental materials (S1 Text). Of those individuals who had visitors before illness, 60% did not receive any visitors when ill, whereas 40% continued to receive visitors during illness. While the majority of participants who did not receive visitors before illness (70%) also did not receive visitors during illness, 30% did gain a visitor during their illness period.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Differences in caregiving and visitor behaviors experienced by symptomatic DENV infected individuals in Iquitos, Peru, compared by sex, age, and Quality of Well-Being (QWB) score.

Fisher’s Exact tests were performed for variables (1)whether caregiving was received, (2)if the person who helped was their housemate, (3) if the person who helped had their work negatively affected, (4) if visitors were received during the illness period, (5) whether one or two visitors were received, and (6) what the relationship of the visitor was (family/friend/other). The percent of each group (and the raw number of participants) that experienced a behavior is listed, as is the p-value for the Fisher’s Exact test. Participants’ minimum QWB score during illness were split into two equal groups, ‘Low’ and ‘High’, where those with ‘Low’ QWB scores had the worse health-related quality of life during illness. Variables (1),(4), and (5) were examined at the participant level, whereas (2),(3), and (6) were examined at the visitor/helper level. For variables (1)-(6) the overall N values in the first 2 columns (sex and age) were 67, 68, 68, 70, 23, and 85, respectively. As there were 2 participants without QWB scores available, overall N values in the QWB score column were 65, 65, 65, 68, 22, and 82, respectively.

https://doi.org/10.1371/journal.pntd.0009614.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Characteristics of visitors received during illness by symptomatic DENV infected individuals in Iquitos, Peru.

Data on visits received by ill individuals and the nature of these visits. Totals and categorized by if made by routine visitors. Whether the visitor knew of the illness, why they visited, and if they were classified as a ‘routine visitor’ were all reported by the ill participant.

https://doi.org/10.1371/journal.pntd.0009614.t002

According to the best-fit GLM, the sex of an ill individual was most associated with whether or not they received visitors during illness (S2 Table). Females were 3.7 times (95% CI: 1.2–10.7) more likely to than males to receive visitors during illness, with predicted probabilities of 50% and 20% for females and males, respectively. When examining the daily likelihood of receiving visitors during illness, the best-fit GLMM by AICc score included the interaction between age and sex (after accounting for participant ID as a random effect), where female children had the highest predicted probability of receiving visitors on any given day of illness (10%) and male children had the lowest daily probability (0.3%) (S3 Table and Fig 1). This model, however, was not significantly better than the GLMM including only sex and random effect of participant when looking at reduction in deviance (χ² Analysis of Deviance, p = 0.08). In this model, females and males had predicted probabilities of 3.2% and 0.5% for receiving visitors on any given day of illness. Accounting for the specific day of illness did not have a significant effect on the fit of the model.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Predicted probability of receiving a visitor on each day of illness, based on a GLMM fitted to data of DENV infected individuals in Iquitos, Peru.

Data are presented in terms of age group and sex. Error bars represent 95% Confidence Intervals.

https://doi.org/10.1371/journal.pntd.0009614.g001

Caregiving behavior

Of the 67 participants who provided data, 60 (90%) had someone help care for them during their illness. Of the seven individuals who did not receive help, five were adult females. Accordingly, children were significantly more likely than adults to receive help (98% vs. 78%; Fisher’s Exact test, p = 0.01) (Table 1). Further, the best-fit model for predicting whether or not caregiving was received included the age of the ill person, where the odds of receiving help was 12.3 times higher for children than adults (95% CI: 1.4–109.7) (S4 Table). There were no significant differences when comparing across age and sex combinations.

For those participants who had caregivers, caregivers most commonly were a relative who lived with them (91%) who helped for the length of the illness (94%). Only 11% of participants received help from two people. According to the best fit model, those who needed personal care help during illness were 5.6 times (95% CI: 1.0–31.5) more likely to receive help from two people compared to those who didn’t need personal care help (S5 Table). Among the people helping sick individuals, 28% had to take time off of work due to their caregiving behavior. It was more likely for work to be negatively affected when helping ill adults compared to ill children (42% vs 18%; Fisher’s Exact test, p = 0.05) and when helping those with low QWB scores compared to those with high QWB scores (39% vs 11%; Fisher’s Exact test, p = 0.01; Table 1). Indeed, the best-fit model for predicting whether a helper’s work would be negatively affected accounted for the participant’s minimum QWB score during illness, which was split into 3 groups of equal size (S6 Table). The predicted probability of having work adversely affected by helping those with the lowest QWB scores was 50%, as compared to a probability of 15% for those with medium QWB scores and 6% for those with high QWB scores. Accordingly, those helping individuals with the lowest QWB scores were more likely to have their work affected compared to those helping individuals with medium (OR: 0.18, 95% CI: 0.05–0.60) and high (OR: 0.07, 95% CI: 0.01–0.60) QWB scores. Additional details on the types of help given by caregivers are provided in the supplemental materials (S1 Text).

Copyright statement

Authors H.A.V., C.N.M., A.C.M., W.L.Q., and E.J.R.L. are or were federal/contracted employee of the United States government. This work was prepared as part of their official duties. Title 17 U.S.C. §105 provides that ‘Copyright protection under this title is not available for any work of the United States Government.’ Title 17 U.S.C. §101 defines a U.S. Government work as a work prepared by a military service member or employee of the U.S. Government as part of that person’s official duties.

Disclaimer

The views expressed in this article reflect the results of the research conducted by the author and do not necessarily reflect the official policy or position of the Department of the Navy, Department of Defense, the National Institutes of Health, or the United States Government.