Data for the present study were derived from the Home Air Filtration for Traffic-Related Air Pollution (HAFTRAP) study, which is a double-blind, randomized crossover trial of in-home HEPA air filtration to reduce UFP exposure in Somerville, MA, USA. The study was approved by the University of Connecticut, School of Medicine IRB. A description of the design and methods has been published [21].

In brief, households were randomized to 30 days of either HEPA filtration or sham filtration, with AP units placed in the living room and bedroom, followed by a 30-day washout period, and then a subsequent 30-day period of the alternative intervention. We installed freestanding, custom-made HealthMate air purifiers manufactured by Austin Air (Buffalo, NY) with or without high efficiency particulate air (HEPA) filters for the sham configuration. Biological measures were collected, and questionnaires were administered at trial entry and subsequently at 30, 60, and 90 days. The primary health outcome for the trial was peripheral systolic blood pressure (SBP), based on our prior findings from a controlled exposure study showing a reduction in SBP with decreased PM. Secondary outcomes were diastolic and central BP and blood biomarkers for inflammation, also based on prior findings [22]. A sample size of 207 participants had been calculated to detect a 2.5-mmHg mean difference in peripheral systolic blood pressure between HEPA and sham filtration [21]. Although air purifiers like the ones used in this study typically cost $500–$1000, there were no purchasing or installation charges for the participants in the trial period. Additionally, to compensate for the cost of electricity consumption, participants received $100 worth of gift cards.

The trial enrolled individuals who were 30 years of age or older, lived full time at their house in Somerville MA within 200 m of the highway, had cognitive capacity to fill out questionnaires, and spoke English or Spanish. Recruitment was by door-to-door canvassing, email lists, outreach at events and referral by previous participants. The trial excluded entry to individuals who currently smoked/vaped or lived with indoor smokers/vapers because air purifiers are ineffective against indoor secondhand smoke. We also excluded those with a history of heart attack, stroke, or other major cardiovascular event and were currently taking anti-hypertensive or anti-inflammatory medications. Each of these health conditions is likely to overwhelm the effects of air pollution on one or more of our health endpoints. Additionally, participants were not enrolled if they had occupational or regular exposures to traffic pollution outside of the home or significant combustion sources inside the home [21]. Data were collected between September and June, with the intent to have participants during cooler months when the levels of ultrafine particulate matter tend to be higher.

Three sources of data were used for the present mixed methods study. Data triangulation involved the primary author comparing and contrasting the findings from across each data source and meeting with the second and fifth author to discuss patterns in the context of the qualitative data. The purpose for triangulating data from three different sources was to cross-check participant responses and therefore increase credibility and validity of the study results [23]. Data sources were given equal weight, rather than prioritizing one source over the other. We sought to tell a holistic story about participant experiences with and reactions to APs. A flowchart of the data collection process is shown in Fig. 1.

Flowchart of study process

The first source of data were responses to follow-up questionnaires administered to all participants at their 30-day and 90-day home visits. The questionnaires contained 35 questions total, but only 11 questions related to the participants’ reactions to the air purifiers, asking how they used the APs during the past 30 days in both their living room and bedroom, and to what extent they were affected by noise produced by the APs. These questions were custom-designed and tailored for this study. Each question is listed in Tables 2 and 3. Data collection occurred monthly and was conducted by a phlebotomist and the project manager of the study.

The second source of data was from a subset of participants who had a HOBO Plug Load Data Logger installed between their air purifier and electrical outlet. Though all participants were informed from the beginning about the different subsets they could opt into during the different months of the study, the subsets were ultimately selected on a participant-by-participant basis, according to participant interest and availability. The purpose of the HOBO Data Logger was to continuously record electricity use to determine when the APs were on versus off, which was then used to calculate the percentage of time the air purifiers were running over the past 30 days.

Finally, structured interviews with a subset of participants were conducted. Again, this subset was selected on a participant-by-participant basis depending on interest and availability. One member of each household enrolled in the study was contacted and invited to participate in a structured qualitative telephone interview. Interviews explored participant motivation for participating in the trial, adherence to trial instructions, and experience with the APs. The interviews lasted an average of 15–20 min. Data collection was continued until saturation was achieved. Responses were recorded by the interviewer in Qualtrics; short answers captured verbatim and read back to the participant to ensure accuracy of the response.

Data were summarized using means with standard deviations for continuous data and were summarized using frequencies with percentages for categorical data. Generalized linear models (GLM) with generalized estimating equations (GEE) were used to compare participant-reported outcomes between the 30-day and 90-day home visits. GEEs, using an exchangeable correlation structure, account for the within-participant correlation between the 30-day and 90-day outcomes. GLMs for continuous outcomes were fit with a normal distribution and identity link, GLMs for ordinal outcomes were fit with a multinomial distribution and cumulative logit link, and GLMs for binary outcomes were fit with a binomial distribution and logit link. All statistical analyses were carried out using SAS 9.4 (SAS Institute Inc., Cary, NC), and results with p-values < 0.05 were deemed statistically significant. Short answers from interviews were exported in Excel, and thematic analysis was employed following the six steps outlined by Braun and Clark (2006) to identify patterns in the data. Coders immersed themselves in the data, developed codes, identified and refined themes, and selected specific examples [24].