PROOF

Smartphone App for Residential Testing of Formaldehyde

(SmART-Form)

Journal: Indoor Air

Manuscript ID Draft

Manuscript Type: Original Article

Date Submitted by the Author: n/a

Complete List of Authors: Zhang, Siyang; Ohio State University, Civil Environmental and Geodetic Engineering Shapiro, Nicholas Gehrke, Gretchen Castner, Jessica Liu, Zhenlei; Syracuse University, Mechanical and Aerospace Engineering

Guo, Beverly; Syracuse University, Mechanical and Aerospace Engineering Satish Prasad , Romesh ; Syracuse University, Mechanical and Aerospace Engineering Zhang, Jianshun; Syracuse University, Mechanical and Aerospace Engineering Haines, Sarah; Ohio State University, Civil Environmental and Geodetic Engineering Kormos, David; Ohio State University, Civil Environmental and Geodetic Engineering Frey, Paige; Ohio State University, Civil Environmental and Geodetic Engineering Qin, Rongjun; Ohio State University, Civil Environmental and Geodetic

Engineering Dannemiller, Karen; Ohio State University, Civil Environmental and Geodetic Engineering; Ohio State University, Environmental Health Sciences

Keywords: Sensor, colorimetric, detection, citizen science, education, exposure

Indoor Air - PROOF

Indoor Air - PROOF

PROOF

1

Smartphone App for Residential Testing of Formaldehyde 1

(SmART-Form) 2

3

4

5

Siyang Zhang1, Nicholas Shapiro

2, Gretchen Gehrke

2, Jessica Castner

3 Zhenlei Liu

4, Beverly 6

Guo4, Romesh Prasad

4, Jianshun Zhang

4, Sarah Haines

5,6,7, David Kormos

6, Paige Frey

8, 7

Rongjun Qin6,9, Karen C. Dannemiller

6,7* 8

9

10

1 Department of Computer Science and Engineering, Ohio State University 11

2 Public Laboratory for Open Technology & Science 12

3 Castner Incorporated 13

4 Department of Mechanical & Aerospace Engineering, Syracuse University 14

5 Environmental Science Graduate Program, Ohio State University 15

6 Department of Civil, Environmental and Geodetic Engineering, College of Engineering, 16

Ohio State University 17 7 Department of Environmental Health Sciences, College of Public Health, Ohio State 18

University 19 8 Strand Associates, Incorporated 20

9 Department of Electrical and Computer Engineering, Ohio State University 21

22

23 *Corresponding author: dannemiller.70@osu.edu, 470 Hitchcock Hall, 2070 Neil Ave, 24

Columbus, OH 43210, 614-292-4031 25

26

27

Page 1 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

2

ABSTRACT 28

Chemical contamination is ubiquitous in the indoor environment, but measurement options 29

are often limited outside of research studies. This is especially true for formaldehyde, a known 30

carcinogen and irritant. The goal of this project was to develop a novel screening tool: a 31

smartphone-based app that can be paired with low-cost colorimetric badges for detection of 32

indoor formaldehyde. The target users include citizen scientists, concerned citizens, public 33

health nurses visiting homes, and researchers with relevant measurement needs. The user 34

interface was designed using a collaborative development model. Badges are exposed to air 35

for 72 hours, and the user takes images that are analyzed in the phone. The app itself measures 36

illumination (lightness) to determine color change, which was associated with formaldehyde 37

concentration (R² = 0.8811, P<0.0001). The detectable range is 20-120 ppb and the standard 38

deviation of readings was 10.9 ppb. Warnings were integrated into the app to address current 39

limitations, including sensitivity to extreme lighting conditions and elevated (>80%) relative 40

humidity. Co-exposure to acetaldehyde or a VOC mixture did not interfere with measurement 41

(P=0.93, P=0.07, respectively). Overall, this screening tool can provide inexpensive, 42

accessible information to users about their formaldehyde exposure, which can inform further 43

testing and potential remediation activities. 44

45

KEYWORDS 46

Sensor, colorimetric, detection, citizen science, education, exposure 47

48

PRACTICAL IMPLICATIONS 49

Smartphone technology can be paired with colorimetric chemistry to design low-cost 50

screening tools for chemical contaminants in the indoor environment. Here, we demonstrate 51

this technology for measurement of formaldehyde and utilize a community-based approach 52

Page 2 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

3

for app development. This process will enhance contaminant measurement options for citizen 53

scientists, concerned members of the general public, and clinicians who conduct home 54

exposure assessments in the future, and enable extensive data collection from many locations 55

compared to other methods. 56

57

INTRODUCTION 58

Formaldehyde is ubiquitous in the indoor environment, where this contaminant is 59

released from various consumer products such as pressed wood, adhesives, paints, cleaners, 60

and more 1. Exposure to this compound can result in eye, nose, and throat irritation, and a 61

range of other non-clinical ailments. Formaldehyde is also a known human carcinogen. 62

Research began demonstrating the toxicity of this compound more than 100 years ago 2. 63

Recent concerns have arisen due to elevated presence in Federal Emergency Management 64

Agency (FEMA) trailers such as those provided after Hurricane Katrina in 2005-2006 and in 65

flooring retailed by Lumber Liquidators in 2015. 66

Formaldehyde measurement options. Currently, measurement options for citizen 67

scientists and concerned members of the general public are limited. A commonly-used method 68

for formaldehyde measurement involves collecting samples in 2,4-dinitrophenylhydrazine 69

(DNPH) cartridges to derivatize formaldehyde with later laboratory-based detection using 70

high performance liquid chromatography 3,4. This technique requires specific sampling 71

equipment, uses expensive and complex analytical instruments, is prone to contamination, and 72

is nearly impossible for a citizen scientist to perform. Other sampling technologies have 73

recently been developed 1,5–9

, but to our knowledge, are not commercially available or widely 74

used by citizen scientists, and may also suffer from contamination either during transport or 75

due to other compounds. Right now, we are missing a low-cost, accessible screening tool for 76

formaldehyde in homes that may not require the same level of precision as the DNPH method. 77

Page 3 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

4

Our previous survey of 147 individuals indicated broad public interest in formaldehyde testing 78

using smartphone technology. Poor reported health status was associated with increased desire 79

for testing (odds ratio for very good to excellent health 0.31, 95% confidence interval 0.12-80

0.81). Location in our targeted area of community engagement was also associated with 81

interest in testing 10. 82

Smartphone-based measurement. Smartphone technologies provide a new 83

opportunity for formaldehyde detection. One downside to the currently available colorimetric 84

badges is the low accuracy due to visible color comparisons by the human eye. This allows 85

for concentration “binning” but not accurate, quantitative results. Fortunately, the use of 86

digital image processing technologies provide the opportunity to more accurately read the 87

color change on the badges that results from formaldehyde exposure as a continuous variable. 88

This is especially important for a compound such as formaldehyde that is ubiquitous in the 89

indoor environment, such that concentration level is more informative than simple detection 90

(yes vs. no). 91

Others have also recognized the opportunity to pair colorimetric sensors with 92

Smartphone detection for accurate measurements 11–14

. In this paired system, the camera 93

function of the smartphone is utilized to record and analyze the color changes on the badge 94

that occur as a result of contaminant exposure. An image formation model describes how the 95

color images are produced by the camera that received light reflected from the badge surface. 96

Colors are dependent upon both the material reflectance and the intensity of the environmental 97

light. It is generally impossible to measure the exact environmental lighting conditions, so the 98

colorimetric change is quantified using a ratio of the lightness (illumination) of badge to a 99

calibration patch 15,16

. Here, we refer to this as the color-change ratio. In this case, the variant 100

of the environment lighting is the same on the reaction and calibration areas, leaving the 101

material property as the only variant accounting for the colorimetric change. This material 102

Page 4 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

5

property, also known as the surface albedo, can be easily computed by analyzing digital 103

photos through extracting the color values of the reacting and referencing area. The surface 104

albedo correlates with the measured formaldehyde concentration. 105

Goal of this work. The goal of the work described here was to develop a novel 106

screening tool for formaldehyde detection by pairing a color-changing badge with a 107

smartphone app (Figure 1). The target user is a citizen scientist or person concerned about 108

formaldehyde exposure in their home environment. The system may also be useful for visiting 109

nurses in asthma homecare programs 17, environmental health professionals, or in certain 110

epidemiological research studies depending on measurement requirements. User input was 111

invited at every stage of development of the low-cost smartphone-based formaldehyde 112

measurement system to ensure that this technology is accessible and useful to the public. 113

Community engagement was sought from both a national network of citizen scientists with 114

high technological literacy and a community environmental group in a small city in southern 115

Georgia with average to low technological literacy. Existing colorimetric badges used in 116

occupational settings were modified by our industry partner for enhanced sensitivity in the 117

residential environment and for an app-friendly configuration. The system was calibrated on 118

both Android and iOS devices using exposure to known concentrations of formaldehyde. 119

Community feedback had been collected previously to assess public interest in residential 120

formaldehyde and home testing, willingness to pay for test kits, smartphone access, and 121

capacity to implement a multi-day sampling protocol 10. Upon completion of a beta version of 122

the app, community members were solicited to provide feedback about the design, clarity, and 123

functionality of this novel system; that feedback was then incorporated into the app. 124

125

Page 5 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

6

METHODS 126

Our goal was to develop a smartphone app to measure changes in color (illumination 127

or lightness) of low-cost colorimetric badges due to exposure to formaldehyde. Steps in this 128

process included badge modification, prototype app development, quantification of color 129

change (also see information on the lighting model in supporting information), system 130

calibration, beta testing with feedback integration, and statistical analysis. 131

Badge modification 132

We used modified SafeAir® formaldehyde detection badges (Morphix Technologies, 133

Virginia Beach, VA) to detect formaldehyde concentrations typically found in the indoor 134

environment. The badges are commonly used in occupational settings and were modified for 135

use with our system. We requested that the badges display a greater intensity of color change 136

for a lower range of formaldehyde exposure concentrations more common in residential 137

environments. We also requested adjustments to the layout of the badges to be more easily 138

read with the app and incorporate a calibration area that does not change color. We also 139

evaluated a modified version of the ChromAir® badges (referred to as “version 2” of the 140

badge), but selected the SafeAir® badges for use with our system due to increased precision 141

and a lower detection limit. 142

Prototype App Development 143

The first steps of app development were to create a back-end image processing method 144

to measure formaldehyde concentration based on substrate color change, and a front-end user 145

interface for the app. The required functionalities included: create multiple tests, take badge 146

image via camera before and after badge exposure, convert badge image into intensity ratio 147

and calculate the concentration result, save multiple tests containing badge images and 148

formaldehyde concentration results, request test data and user experience feedback from users 149

and incorporate relevant changes. 150

Page 6 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

7

Prototype. The first step of app design was creating the framework of our prototype. 151

To facilitate organizing multiple tests, the main layout was designed with a master-detail two-152

panel structure: table view and detail view. The table view contains all existing tests with 153

titles and dates, and the detail view displays the whole process of a badge test, including user 154

input parameters, images taken by the camera, and access to data upload surveys (Figure S1). 155

Our design strategy was closely related to the functional requirements and utilized object-156

oriented design, which is the process of planning a system of interacting objects. The object-157

oriented design structure was focused on the badge test, and for each test, we included the 158

unique ID, title, date, images as inputs, and result as output. For memory efficiency, the test 159

objects were stored in JSON format and images were referred by path in the internal storage. 160

The main functional requirements fall into two categories: test management and image 161

processing. As shown in Figure 2, the app allowed the user to create a new test by inputting 162

the test title, and the start time was recorded after taking the “before” image. Afterward, the 163

user has access to view all existing tests via the table view and check details by clicking a test 164

in the list. For camera setup, the original plan was to use the default camera embedded with 165

the smartphone. However, default parameters vary by phone manufacturer and model, and 166

many of the smartphone cameras in different systems are rather restrictive in the customizable 167

function settings (e.g. ISO level, aperture control). Thus it is challenging to realize 168

homogeneity in terms of colors and lighting sensitivity across different smartphone cameras in 169

different systems. Therefore, in our app, we experimented with taking images with different 170

parameters on different phones, and selected a set of the most common parameters each for 171

Android and IOS systems to minimize variation (see Lighting Experiments section below). 172

Our mathematic model uses the calibration patch on the badge to account for environmental 173

variation, and thus we expect the differences in camera parameters can be partially addressed. 174

Page 7 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

8

For details on the mathematical model used, please refer to the supporting information (Figure 175

S1). 176

User Interface design. The initial user interface for the detail page contained multiple 177

functionalities, such as taking a photo of the surroundings of the test, taking pre-exposure 178

(“before”) and post-exposure (“after”) images, and uploading data to a survey platform 179

(Qualtrics). Surveys were intended to engage users in creating a database of formaldehyde 180

exposure information, and these included a survey about personal health and a survey about 181

environmental characteristics of the sampling location. A unique non-identifiable code is 182

automatically copied to user phone's clipboard when leaving the app and can be pasted into 183

the ID field in the survey, which allows multiple survey and formaldehyde test entries to be 184

linked together without additional personal identifying information. 185

186

Quantification of color change 187

Integration of user warnings. We identified key conditions that could lead to 188

erroneous readings throughout app development (Table 1). For each condition, we analyzed 189

images taken under both suitable (no error) and unsuitable (may cause an error) conditions. 190

The parameter associated with each error, such as lightness or saturation, was chosen based on 191

the value with the most substantial difference between conditions. Boundaries for warnings 192

were selected to be about halfway between extreme points measured in suitable and unsuitable 193

conditions. 194

Lighting experiments. We also conducted lighting experiments to determine limits on 195

lighting conditions in the indoor environment for accurate calculation of illumination values 196

for the badge. When developing this application, we needed to take clear images with no 197

shadows and correct color composition. We also tested the accuracy of the automatic camera 198

settings on the iOS and Android phones and compared to alternative settings. To do this we 199

Page 8 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

9

purchased different types of lightbulbs and a desk lamp, setting up the lamp 30.5 cm from the 200

badge with the lampshade horizontal to avoid shadows covering the badge. The types of 201

lightbulbs purchased were: 25W Soft White, 40W Clear, 40W LED, 50W Soft White, 60W 202

Crystal Clear and 60W LED Soft White. 203

We photographed the color grid paper under three different manual camera settings 204

and auto adjust. Four settings (Table S2) were tested using special color grid paper (Figure 3). 205

After each image was taken, they were imported into Matlab and cropped by reaction area and 206

calibration area. These cropped bitmaps from the orange and yellow blocks were converted 207

into HSI (Hue, Saturation, and Intensity) color model 18. We found that the color-changing 208

reaction of the badge changes the color lightness. Lightness is correlated with the intensity 209

value of the HSI transformation, so we then calculate the ratios of the means of color intensity 210

(I) of each image. We used the color grid paper instead of the actual badge because the printed 211

color will not change, thus providing a good reference to determine the influence caused by 212

the camera settings. 213

Algorithm. The mathematical model is based on the color change ratio (illumination 214

or lightness) between the calibration patch and the chemical badge. Figure 4 highlights the 215

areas in green and red, respectively. These two blocks of bitmap were cropped and calculated 216

for the mean value of the color intensity I, and the ratio = I(red)/I(green). We confirmed a 217

linear relationship between exposure (formaldehyde concentration in ppb·hr) and the color 218

change ratio. 219

220

Calibration 221

Calibration was performed by subjecting the badge sensor to several known 222

concentrations of formaldehyde maintained in a 50 L small-scale environmental chamber (0.5 223

m by 0.4 m by 0.25 m, Figure S2). The test system included a pressurized clean air supply and 224

Page 9 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

10

a Dynacalibrator containing a permeation tube at a constant temperature and flow rate to 225

provide the desired formaldehyde generation rate (Figure S3). Different concentrations in the 226

chamber were obtained by adjusting the total airflow rate through the chamber, and were 227

verified by air sampling with DNPH cartridge followed by HPLC analysis (Table 2). For each 228

test, a set of three badge sensors were placed inside the chamber (Figure S4), and their color 229

change was measured over time. Images were taken before the badges were placed in the 230

chamber, after exposure inside the chamber over the different time periods through a pane of 231

glass, and after they were removed from the chamber by using the mobile phone app (Huawei 232

Mate 8 Android 7.0 and iPhone 6s iOS 11.2) and a high resolution digital camera (Casio 233

Exilim- F1). Data were analyzed to obtain the mean color change ratio of the three badges as a 234

function of the amount of cumulative exposure (i.e., concentration multiplied by the 235

cumulated exposure time, (E = C × T): CCR = �() = �(� × �), where CCR is the color 236

change ratiom, E is total exposure, C is formaldehyde concentration, and T is time. Given this 237

calibration curve, the concentration can be determined from the color change ratio and 238

exposure time. The uncertainty of the badge response was quantified by the standard deviation 239

of the color change ratio measured by the three badge sensors. Taking images through the 240

glass had a small effect on the measured values, so the mathematical model was adjusted for 241

this “glass effect” by comparing images taken without the glass (out of the chamber) and 242

through the glass (in the chamber) at the same time point, both before and after testing. 243

244

245

Beta Testing 246

We conducted a user test of the app’s beta version to ensure that the SmART-Form 247

app is useful for and accessible to interested members of the public. We followed the 248

principles of user-centered design 19. We asked participants to download and use the app with 249

Page 10 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

11

an image of an exposed badge, and to provide feedback on the app’s design, clarity, and 250

functionality. Participants provided feedback to study staff through a survey in Qualtrics with 251

questions about whether or not they had trouble at each step of the app, from initial 252

downloading through survey completion, and were asked for general feedback and 253

suggestions. We recruited beta test participants through: (1) email solicitation of people who 254

had participated in the initial feasibility assessment study, (2) social media solicitation of an 255

environmental community science network, (3) solicitation on the project website. 256

Recruitment targeted both a national network of citizen scientists with high technological 257

literacy and a community environmental group in a small city in southern Georgia with 258

average to low technological literacy. One purpose of the beta test was to evaluate image 259

collection on different smartphone systems. Another core component of the beta test was to 260

investigate whether or not the intended flow of the app, including exiting the app in order to 261

provide health and environmental data through a secure browser-based platform Qualtrics, 262

was intuitive, confusing, or caused technical challenges for users. 263

The user feedback portion of the beta test was conducted through an analogous survey 264

on Qualtrics. The survey was available from September 2017 to May 2018. Feedback was 265

then integrated into app design. To support improvements, a user interface consultant was 266

hired to provide guidance, as the usability of the app is of fundamental importance. We also 267

tested our app in a field test that will be reported in a future publication and solicited some 268

verbal user feedback there. 269

This study was approved by the Ohio State Institutional Review Board (IRB). 270

271

Statistical analysis. 272

Statistical analysis was conducted in SAS, version 9.4 and Microsoft Excel. 273

Formaldehyde concentrations were evaluated as ppb·hr and badge color change was evaluated 274

Page 11 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

12

as the illumination ratio of the color changing area to the calibration area (referred to 275

throughout as the color change ratio). The method detection limit was calculated as 3 times 276

the standard deviation of blank readings. The standard deviation of all the data was calculated 277

by comparing the actual formaldehyde concentration to the calculated formaldehyde 278

concentration based on the color change ratio and our mathematical model. Cross-279

contamination of acetaldehyde and VOCs were evaluated using the GENMOD procedure in 280

SAS and considering the interaction term based on grouping reading based on presence or 281

absence of acetaldehyde and using the “no intercept” option. 282

283

RESULTS 284

Our testing of the badges demonstrated the ability of smartphone cameras to accurately 285

determine changes in illumination (color change) associated with formaldehyde exposure. 286

287

App Development 288

The app user interface was developed and refined using a collaborative development model to 289

allow both measurements of the color change of the badges and access to surveys that can be 290

made publicly available (Figure 3,4). Modifications were made to the user interface 291

throughout the app design process (Table 3). 292

293

Necessary user warnings and education 294

Integration of user warnings. The image output relies on the camera settings and the 295

lighting conditions, and it is important to reduce the possibility of overexposure or low 296

lighting conditions. To avoid such test errors, we identified a list of warnings to test and 297

subsequently notify the user to instruct them on the appropriate use of the app (Table 1). 298

Page 12 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

13

The values in Table 1 were determined from a series of experiments on the badges. 299

Different experimental conditions were tested, and boundaries for warnings were selected to 300

be about halfway between extreme points measured in suitable and unsuitable conditions. 301

For instance, we noted that high relative humidity conditions above about 75-80% will 302

interfere with the color change of the badge and can potentially cause erroneous readings in 303

the app. Fortunately, under these conditions the badge also develops a blue/purple tint that is 304

detectable in the image. We incubated badges at various relative humidity conditions to 305

determine when an interfering blue color appeared on the badge. The lowest saturation value 306

detected in suitable conditions was 0.8008 and the highest saturation value detected in 307

unsuitable conditions was 0.5185. Therefore, we selected 0.6 as the boundary for activation of 308

this warning. We also subjected badges to long-term, high-temperature storage conditions to 309

produce contamination and quantified detection with the app. 310

For exposure time, we selected a 72 hour period to balance detection capabilities with 311

user time and consistency in the color-changing area of the badge. The badge can theoretically 312

be read at different times that allow for sufficient color change (at least about 12 hours), but 313

those parameters were not validated here. Taking an image in a very short amount of time (for 314

instance, 5 minutes) will result in an artificially high value by dividing by a very small time 315

value due to the algorithm used, and thus we also wanted to prevent this error. We also placed 316

a limit on the reported values so that high and low values will be reported as >120 ppb and 317

<20 ppb, respectively. This range represents the calibration range above the method detection 318

limit. Values above 120 ppb may have also experienced saturation on the badge, but this 319

needs further evaluation to confirm. 320

User education. We integrated a formaldehyde information sheet into the app, with 321

the intention of providing necessary information to users to interpret their results and take 322

Page 13 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

14

appropriate action if necessary. The informational page lists typical formaldehyde 323

concentrations and potential sources and remediation strategies. 324

325

Camera Parameters and Lighting Conditions. Our results at different lighting 326

conditions and camera settings informed our decisions related to app design. Overexposure 327

was a challenge that could potentially result in inaccurate readings, but fortunately, we could 328

prevent these inaccuracies by incorporating warning messages into the app. Under our 329

experimental conditions with a close light source 30.5 cm from the badge, the ratio of color 330

intensity between “orange” and “yellow” is quite stable with different camera settings under 331

25W, while the results of 60W show more variations (Figure S5). In our setup, the 25W or 332

40W non-LED light bulb worked best (Figure S6), and we were able to determine the optimal 333

bounds for lighting conditions. The variations are possibly caused by overexposure as well, so 334

the ideal camera settings should eliminate such overexposure. We also determined that it was 335

necessary to use fixed camera settings as opposed to automatic settings, which vary by phone. 336

We selected White Balance: 4000K, ISO: 200, Exposure Duration: 1/60s or 1/125s depending 337

on external lighting conditions, to avoid any overexposure (Table S2). Other uncontrolled 338

camera settings may cause small differences in image quantifications, especially under 339

extreme lighting conditions. However, we expect the system to perform well under most 340

“normal” indoor lighting situations. We incorporated error messages into the app to alert the 341

user if lighting conditions exceed acceptable bounds, allowing them to take a better picture for 342

more accurate results. 343

344

Calibration 345

The results of color change ratio as a function of exposure time, CCR = f(t), for the 346

three different levels of formaldehyde concentrations under the same temperature (23 °C) and 347

Page 14 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

15

relative humidity (50% RH) are shown in Figure S8. The badge sensors did not appear to 348

suffer from saturation up to the 72 hours tested. The sensors of the low concentration test 349

showed only minimal color change at 72 h because this was close to the limit of detection. In 350

order to test the limit of sensor under the low indoor concentration level, we extended the test 351

of the low concentration test to 360 h until it reached the saturation state. The results show 352

that the function CCR = f(t) shows the trend of linear relation for all the three levels of 353

formaldehyde exposure conditions (Figure S8). The results also show that the color change of 354

version1 (SafeAir®, Morphix Technologies, Virginia Beach, Virginia) of the badge sensor is 355

more sensitive to the formaldehyde exposure than version2 (ChromAir®, Morphix 356

Technologies, Virginia Beach, Virginia). 357

The results of high relative humidity test are plotted in Figure S9. Under high RH 358

condition, the CCR value increased instead of decreased with time, indicating a RH 359

interference. This was partially due to a blue/purple tint that occurs under this condition. The 360

CCR at the high RH also varied with time or exposure amount, but in a non-linear fashion. It 361

is therefore recommended not to use the current sensor at high RH, and a warning was 362

integrated into the app that detects this blue/purple color. 363

Figure 7 shows the exposure as a function of CCR and time for all the tests conducted 364

at 23 C and 50% RH without potential contaminants, including a linear correlation function 365

that was implemented in the app. Illumination (lightness) from the color change was 366

associated with formaldehyde concentration (R² = 0.8811, P<0.0001). 367

We calculated the method detection limit to be 20 ppb at 72 hours. We also placed an 368

upper limit of 120 ppb detection due to concerns about saturation of the reaction area beyond 369

what was tested. The standard deviation of the data was calculated to be equivalent to 10.9 370

ppb if measured at 72 hours of exposure. Assuming a normal distribution, we expect 68% of 371

Page 15 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

16

the measured values to fall within 10.9 ppb of the true value at 72 hours of exposure, and 95% 372

of the data will be within 21.8 ppb of the true value. 373

The data in Figure 7 are from the Android system, and a similar graph with nearly 374

equivalent linear model resulted from the DLSR camera images. The iPhone that was used in 375

testing developed an artificial blue tint only when images were taken through the glass that 376

interfered with analysis, and thus the Android results were used for all versions of the app. 377

Further testing indicated that Android and iOS systems yielded similar results within error 378

limits when imaging the same badge under appropriate lighting conditions (P=0.70). We do 379

not expect the blue tint to interfere in future use of the app on iOS systems because 1) it 380

appeared to be limited to the phone used and 2) we do not expect users to be taking images 381

through glass under normal circumstances. 382

We also tested the badges under two additional conditions at the mid-concentration 383

level to examine the possible interface effect of the co-existence of acetaldehyde, which is 384

another major VOC contaminant in buildings, and other VOCs from the composite wood 385

product source such as a particleboard. The results were not statistically different with and 386

without co-exposure to acetaldehyde (P=0.93) or with and without co-exposure to a VOC 387

mixture from the particleboard (P=0.07) (Figure 7B). 388

Beta Testing 389

We received 10 responses to our beta test survey that both consented and completed at 390

least one question on the survey. Four users reported their age with a mean age of 33 years. 391

We received three other responses where users did not consent to the consent form, and we 392

are unable to determine how many people in total attempted the beta test. Users rated the 393

usability of the app at the time of testing with a mean of 8.1 on a 1 (difficult) to 10 (easy) 394

scale. Users cited “ease of use,” “simple use,” and “clean layout and intuitive” as the main 395

strength of the app (Table 4). The biggest challenge occurred with accessing the survey. 396

Page 16 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

17

Importantly, beta testers reported a bug that impeded the Unique ID assigned to their app from 397

being automatically copied to the Qualtrics survey, which is essential for contextual data 398

collection while protecting personally identifiable information. Because of this issue, many 399

users had difficulty submitting their feedback within the beta test survey and 9 additional beta 400

testers emailed feedback directly to research team members. Other more minor technical 401

glitches included compatibility with down market and older phones. These glitches were 402

addressed to ensure the app functioned properly. 403

Another suggestion received from three beta testers and others who reviewed the app 404

was the need for more clear instructions, both with an overview instruction page and step-by-405

step instructions for each phase. The user interface was substantially transformed following 406

beta testing to address the app’s clarity by adding instructional pages, improving the screen 407

layout to emphasize key information, utilize familiar icons, and ensuring more intuitive user 408

pathways. 409

Overall, the app and beta test had global appeal. There were 136 app downloads from 410

September 2017 to May 2018, although a small but unknown number of those included from 411

the study team. Country of download included the United States, India, South Korea, South 412

Africa, Mexico, China, Italy, Canada, United Kingdom, Turkey, Afghanistan, Brazil, Czech 413

Republic, Chile, Germany, Japan, Namibia, Malawi, Niger, Taiwan, Ukraine, and Russia. 414

415

DISCUSSION 416

We have developed a novel measurement system for formaldehyde in the indoor 417

environment. Our current results indicate that this system can provide accurate results and is 418

amenable to use by citizen scientists and concerned members of the general public, as well as 419

visiting nurses and researchers 10,20

. Additional field testing in a community showed color 420

change of the badge and measurement by the system, and will be described in detail in a 421

Page 17 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

18

future publication. Eventually, this technology, which is openly-licensed and open sourced 422

(https://github.com/publiclab/SmART-Form), can be expanded to detect additional 423

compounds of concern in the indoor environment. Integration into a smartphone app also 424

provides an important venue for user education. 425

Formaldehyde is present in most indoor environments 1,4,21,22

, and is notoriously 426

difficult to measure. The contaminant is too volatile to effectively adhere to many common 427

sorbents, and also requires derivitization for detection by gas chromatography/mass 428

spectrometry. Ubiquitous presence also increases the possibility for ambient contamination 429

during sampling. 430

Challenges. Uncontrolled lighting conditions present the greatest challenge for use of 431

this system. Shadows, overexposure, multiple light scattering, and low light are situations that 432

could affect the reading of the badge. Previous work has considered these challenges in 433

applications of surface material and optical property measurement. Normally, measuring the 434

albedo of a surface requires a lab-based reflectance measurement 23,24

, where a single light 435

source and an object with known shape (i.e. a perfect sphere) in a dark room. In a natural or 436

indoor environment, surface albedo measurement requires physical-model based disturbance 437

correction (e.g. atmosphere and light scattering) with in situ information such as scene 438

geometry (walls, tables, etc. in the room) or humidity/weather conditions 25–27

. In our system, 439

we know that the badge is a flat surface, while the environment and its lighting are 440

unpredictable. We designed the badge to include a calibration patch, where we assumed that 441

this portion in the smartphone image encodes variants of the environment. Using this, we are 442

able to approximate the albedo computation through calculating the ratio of lightness of the 443

reaction and calibration areas. We also integrated warnings to the user within the app to 444

account for some of these conditions, but potential still remains for introduced error. Future 445

Page 18 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

19

enhancements of the system can focus on further improving mathematical models used for 446

color calculation, or potentially improving conditions under which images are taken. 447

The standard deviation of our data was 10.9 ppb if the badge has been exposed for 72 448

hours. For instance, this means that a reading of 35 ppb is 68% likely to have a true value 449

between 24-46 ppb, and a reading of 85 ppb is 68% likely to have a true value between 74-96 450

ppb. It is always desirable to obtain a more precise reading. However, the precision available 451

here is most likely acceptable to citizen scientists or concerned citizens who want to quickly 452

determine the general range of their formaldehyde exposure with an inexpensive and easily-453

accessible method. This device is best used as a screening tool to determine if their exposure 454

is high or low, and can inform decisions about further testing or potential remediation. 455

Future use and interventions. This system is not as precise as the gold-standard 456

DNPH cartridge sampling and HPLC analysis method 28, but it does serve as an important 457

screening tool for formaldehyde in the environment. By comparison, the DNPH cartridge 458

method has a reported coefficient of variation of 1.02% and a method standard deviation of 459

0.71 µg/m3 (approximately 0.57 ppb)

29. Such extreme precision is not necessary in all 460

situations, especially when other factors such as ventilation changes or occupant activity may 461

alter formaldehyde levels over different periods of time. An interested user will need to weigh 462

cost, usability, accessibility, and accuracy/precision when deciding which measurement tool 463

to employ. In some cases, the smartphone-based system described here may provide the best 464

combination of these factors, and also provides additional information to the user related to 465

interpretation of results and options for remediation. After use, the user can decide to either 1) 466

pursue more expensive/difficult but more precise testing, 2) conduct additional tests with the 467

SmART-Form system, 3) do nothing, or 4) take remedial action to reduce exposure. More 468

precise and more expensive/cumbersome methods can be employed if desired. Additionally, 469

future improvements can potentially increase precision in future iterations of this system. 470

Page 19 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

20

The use of a smartphone for measurement of environmental exposures presents a 471

novel opportunity for user education and engagement. Increasing viability of low-cost and 472

easy-to-use monitoring systems is already changing the current paradigm of air quality 473

monitoring, according to the U.S. Environmental Protection Agency and others. The historical 474

monitoring model relied on highly expensive and complex instruments with limited input 475

from the public on where they are deployed and limited access of the data once collected. A 476

new paradigm that is variously influenced by citizen science, participatory civic engagement, 477

and the popularity of ‘quantified self’ technologies is seen to bear the possibility of bringing 478

communities and members of the general public into closer dialog with scientists about air 479

quality, enhancing source compliance monitoring and yielding more robust understandings of 480

personal exposures 30,31

. Most of the “next generation” air quality monitors are electronic real-481

time sensors that pose ongoing calibration and drift issues, and are relatively inexpensive (150 482

USD or more) but still too expensive for the communities that bear the highest burdens of 483

environmental exposure. Even highly motivated and scientifically literate users are finding 484

that they are “drowning in data” due to the large volumes of data produced by real time 485

sensors 32,33

. For smartphone colorimetric systems, the analytical technology is already owned 486

by 77% of the U.S. population (with higher rates in some Asian, Middle Eastern, and 487

European countries), allowing a truly accessible price point with badges costing 488

approximately 5 USD and the app being free. Further, these badges represent discrete 489

averages of exposure that are more easily understandable, meaningful, and actionable for the 490

general public than thousands of real-time data points. Information in the app can also assist 491

users in result interpretation and decision-making related to remediation. 492

Fortunately, there are interventions available that can reduce formaldehyde exposure 493

once it is detected. These include primarily source removal or avoidance, but may also 494

involve increasing ventilation rates 1. Future work might explore how occupants respond to 495

Page 20 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

21

detecting high levels of formaldehyde with this app, and if they are likely to take action to 496

reduce their exposure. 497

Limitations. Limitations of our system include potential error introduced due to 498

uncontrolled and potentially untested lighting conditions. We expect our system to function 499

properly under most indoor lighting conditions and provide a warning to the user when 500

lighting conditions are outside of system requirements. However, it is still possible that 501

particular shadows or other untested conditions could affect results. We tested the badges 502

under various lighting conditions, but the lightbulbs in our test were placed close to the badge 503

(30.5) and this will generally not reflect the typical light distance in a given room. Thus, the 504

bulbs used are reported here for reproducibility, but may not indicate the bulbs that are 505

necessary to use in a room where the light source is placed further away from the badge, 506

Additionally, error in the final reading could potentially be reduced with future badge 507

enhancements such as by controlling the position in which the user takes the image, taking 508

multiple images from which to calculate the final value, and further refining our mathematical 509

model. We tested and did not see interference from co-exposure to acetaldehyde or other 510

compounds present in pressed wood products. However, it is also possible that other untested 511

compounds, such as acetone or ammonia, may cause interference in the color change of the 512

badge. Our system is not able to be used under conditions of elevated relative humidity above 513

about 75-80%, which causes an artificial blue tint to the color changing area of the badge. 514

Additional work needs to be done to ensure sufficient communication of measurement 515

precision and error within the app. Further development and testing of this technology may be 516

able to overcome some of the limitations listed here. 517

Strengths. Strengths of this study include the development of a novel screening tool to 518

use a smartphone and color-changing badge to measure formaldehyde concentration in the air. 519

This system was developed in conjunction with end-users across the spectrum of 520

Page 21 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

22

technological literacy to expand potential use. We found a strong correlation between 521

exposure to formaldehyde and color change in the badge as read by the smartphone camera. 522

Thus, the badge is useful as an inexpensive indicator that formaldehyde is present and 523

detectible, while additional modification is required to enhance the precision of the 524

quantitative results. Given the wide range of plausible values from each reading, and levels of 525

uncertainty, we recommend use by well-informed users, professionals, or paraprofessionals 526

with risk communication training to interpret results and recommended actions until greater 527

quantitative precision can be achieved. However, information in the app can also assist with 528

interpretation of results and user education. This app is currently available for download on 529

both Android and iOS devices. This system can be further improved upon to provide results 530

that are more precise in the future. Our system makes formaldehyde measurement more 531

accessible to community scientists and other interested uses and also provides a new 532

opportunity for both scientific engagement and education. 533

534

CONCLUSIONS 535

Here, we developed a novel system for smartphone-based formaldehyde measurement 536

system for testing of residential formaldehyde levels. This new method can expand 537

measurement options for a range of future users and potentially also be applied to additional 538

compounds of interest. This current system is best used as a screening tool to determine the 539

general level of formaldehyde in an environment when cost, usability, and accessibility are the 540

most important considerations. We expect that future improvements to this and similar 541

systems can improve precision. The use of a smartphone encourages engagement of 542

community scientists and concerned citizens, and also provides an additional opportunity for 543

education about formaldehyde exposure. This platform can also be used by researchers in 544

epidemiological studies as well as medical providers conducting home visits for vulnerable 545

Page 22 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

23

populations, such as asthmatics. Finally, this system can be used to respond to future 546

widespread concerns that might occur related to formaldehyde exposure. 547

548

ACKNOWLEDGEMENT 549

Funding was provided by National Science Foundation (NSF) grant 1645048. Content has not 550

been reviewed by NSF. We also thank our industry collaborator, Morphix Technologies, for 551

manufacturing and providing novel home testing badges for the research team’s testing in this 552

study. We also want to acknowledge the community scientists who have contributed to the 553

project. A special thanks to Joan Tibor for her efforts related to use of this app. The authors 554

are grateful for user interface design advice from Kevin Nguyen, who also provided the 555

wireframe diagram included in this manuscript. 556

557

558

REFERENCES 559

1. Salthammer T, Mentese S, Marutzky R. Formaldehyde in the Indoor Environment. 560

Chem Rev. 2010;110(4):2536-2572. doi:10.1021/cr800399g. 561

2. Kersey RW, Maddocks JR, Johnson TE. The determination of small amounts of 562

formaldehyde in air. Analyst. 1940;65(769):203-206. doi:10.1039/AN9406500203. 563

3. Sexton K, Liu KS, Petreas MX. Formaldehyde concentrations inside private residences: 564

a mail-out approach to indoor air monitoring. J Air Pollut Control Assoc. 565

1986;36(6):698-704. 566

4. Grosjean D. Ambient levels of formaldehyde, acetaldehyde and formic acid in southern 567

California: results of a one-year baseline study. Environ Sci Technol. 1991;25(4):710-568

715. doi:10.1021/es00016a016. 569

5. Feng L, Musto CJ, Suslick KS. A Simple and Highly Sensitive Colorimetric Detection 570

Method for Gaseous Formaldehyde. J Am Chem Soc. 2010;132(12):4046-4047. 571

doi:10.1021/ja910366p. 572

6. Rocha FR, Coelho LHG, Lopes MLA, et al. Environmental formaldehyde analysis by 573

active diffusive sampling with a bundle of polypropylene porous capillaries followed 574

by capillary zone electrophoretic separation and contactless conductivity detection. 575

Talanta. 2008;76(2):271-275. doi:10.1016/j.talanta.2008.02.037. 576

7. Suzuki Y, Nakano N, Suzuki K. Portable Sick House Syndrome Gas Monitoring 577

System Based on Novel Colorimetric Reagents for the Highly Selective and Sensitive 578

Detection of Formaldehyde. Environ Sci Technol. 2003;37(24):5695-5700. 579

doi:10.1021/es0305050. 580

8. Wang X, Si Y, Wang J, Ding B, Yu J, Al-Deyab SS. A facile and highly sensitive 581

Page 23 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

24

colorimetric sensor for the detection of formaldehyde based on electro-spinning/netting 582

nano-fiber/nets. Sensors Actuators B Chem. 2012;163(1):186-193. 583

doi:http://dx.doi.org/10.1016/j.snb.2012.01.033. 584

9. Chung P-R, Tzeng C-T, Ke M-T, Lee C-Y. Formaldehyde Gas Sensors: A Review. 585

Sensors (Basel). 2013;13(4):4468-4484. doi:10.3390/s130404468. 586

10. Castner J, Gehrke GE, Shapiro N, Dannemiller KC. Community interest and feasibility 587

of using a novel smartphone-based formaldehyde exposure detection technology. 588

Public Health Nurs. January 2018. doi:10.1111/phn.12384. 589

11. Hong J Il, Chang B-Y. Development of the smartphone-based colorimetry for multi-590

analyte sensing arrays. Lab Chip. 2014;14(10):1725-1732. doi:10.1039/c3lc51451j. 591

12. Jarujamrus P, Meelapsom R, Pencharee S, et al. Use of a Smartphone as a Colorimetric 592

Analyzer in Paper-based Devices for Sensitive and Selective Determination of Mercury 593

in Water Samples. Anal Sci. 2018;34(1, SI):75-81. doi:10.2116/analsci.34.75. 594

13. Solmaz ME, Mutlu AY, Alankus G, Kılıç V, Bayram A, Horzum N. Quantifying 595

colorimetric tests using a smartphone app based on machine learning classifiers. 596

Sensors Actuators B Chem. 2018;255:1967-1973. doi:10.1016/J.SNB.2017.08.220. 597

14. Ra M, Muhammad MS, Lim C, Han S, Jung C, Kim W-Y. Smartphone-Based Point-of-598

Care Urinalysis Under Variable Illumination. IEEE J Transl Eng Heal Med. 2018;6:1-599

11. doi:10.1109/JTEHM.2017.2765631. 600

15. Basri R, Jacobs DW. Lambertian reflectance and linear subspaces. IEEE Trans Pattern 601

Anal Mach Intell. 2003;25(2):218-233. doi:10.1109/TPAMI.2003.1177153. 602

16. Barmpoutis A, Kumar R, Vemuri BC, Banerjee A. Beyond the Lambertian assumption: 603

A generative model for Apparent BRDF fields of faces using anti-symmetric tensor 604

splines. In: 2008 IEEE Conference on Computer Vision and Pattern Recognition. 605

IEEE; 2008:1-6. doi:10.1109/CVPR.2008.4587770. 606

17. Paranzino GK, Butterfield P, Nastoff T, Ranger C. I PREPARE: development and 607

clinical utility of an environmental exposure history mnemonic. AAOHN J. 608

2005;53(1):37-42. http://www.ncbi.nlm.nih.gov/pubmed/15675156. Accessed June 20, 609

2018. 610

18. Schwarz MW, Cowan WB, Beatty JC. An experimental comparison of RGB, YIQ, 611

LAB, HSV, and opponent color models. ACM Trans Graph. 1987;6(2):123-158. 612

doi:10.1145/31336.31338. 613

19. Shah SGS, Robinson I, AlShawi S. Developing medical device technologies from 614

users’ perspectives: A theoretical framework for involving users in the development 615

process. Int J Technol Assess Health Care. 2009;25(4):514-521. 616

doi:10.1017/S0266462309990328. 617

20. Amiri A, Pryor E, Rice M, Downs CA, Turner-Henson A, Fanucchi M V. 618

Formaldehyde Exposure During Pregnancy. MCN, Am J Matern Nurs. 2015;40(3):180-619

185. doi:10.1097/NMC.0000000000000125. 620

21. Dannemiller KC, Murphy JS, Dixon SL, et al. Formaldehyde concentrations in 621

household air of asthma patients determined using colorimetric detector tubes. Indoor 622

Air. 2013;23(4). doi:10.1111/ina.12024. 623

22. Gilbert NL, Gauvin D, Guay M, et al. Housing characteristics and indoor 624

concentrations of nitrogen dioxide and formaldehyde in Quebec City, Canada. Environ 625

Res. 2006;102(1):1-8. doi:10.1016/j.envres.2006.02.007. 626

23. Biswas S, Aggarwal G, Chellappa R. Robust estimation of albedo for illumination-627

invariant matching and shape recovery. IEEE Trans Pattern Anal Mach Intell. 628

2009;31(5):884-899. 629

24. Szeliski R. Computer Vision: Algorithms and Applications. Springer-Verlag London; 630

2010. 631

Page 24 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

25

25. Gordon H, optics MW-A, 1994 undefined. Retrieval of water-leaving radiance and 632

aerosol optical thickness over the oceans with SeaWiFS: a preliminary algorithm. 633

osapublishing.org. 634

26. Kaufman YJ, Tanré D, Dubovik O, Karnieli A, Remer LA. Absorption of sunlight by 635

dust as inferred from satellite and ground-based remote sensing. Geophys Res Lett. 636

2001;28(8):1479-1482. doi:10.1029/2000GL012647. 637

27. Tucker CJ, Sellers PJ. Satellite remote sensing of primary production. Int J Remote 638

Sens. 1986;7(11):1395-1416. doi:10.1080/01431168608948944. 639

28. ASTM International. ASTM D5197 - 16 Standard Test Method for Determination of 640

Formaldehyde and Other Carbonyl Compounds in Air (Active Sampler Methodology). 641

2016. doi:10.1520/D5197-16. 642

29. Sandner F, Dott W, Hollander J. Sensitive indoor air monitoring of formaldehyde and 643

other carbonyl compounds using the 2,4-dinitrophenylhydrazine method. Int J Hyg 644

Environ Health. 2001;203(3):275-279. doi:10.1078/s1438-4639(04)70038-3. 645

30. Snyder EG, Watkins TH, Solomon PA, et al. The Changing Paradigm of Air Pollution 646

Monitoring. Environ Sci Technol. 2013;47(20):11369-11377. doi:10.1021/es4022602. 647

31. Piedrahita R, Xiang Y, Masson N, et al. The next generation of low-cost personal air 648

quality sensors for quantitative exposure monitoring. Atmos Meas Tech. 649

2014;7(10):3325-3336. doi:10.5194/amt-7-3325-2014. 650

32. Ottinger G, Zurer R. Drowning in Data. Issues Sci Technol. 2011;27:71-82. 651

doi:10.2307/43315490. 652

33. Shapiro N, Kirksey E. Chemo-Ethnography: An Introduction. Cult Anthropol. 653

2017;32(4):481-493. doi:10.14506/ca32.4.01. 654

655

656

657

658

TABLES 659 660

Table 1. SmART-Form app warning list. Boundary levels are listed in Table S1. 661

Warning Type Condition Solution Suggested to User

Overexposure Average lightness > 0.8 Retake the badge photo with less light

Low-light Average lightness < 0.4 Retake the badge photo with more light

Badge Contamination Lightness ratio > 1.0 Redo the test with a new badge

Blue tint due to high

relative humidity

Average saturation < 0.4 Redo the test with a new badge

Short exposure time Exposure time < 12 hours Wait for 72 hours of total exposure

Below detection limit Concentration < 20 ppb Concentration too low to detect

Above detection limit Concentration > 120 ppb Concentration higher than detectable

limit

662

663

664

665

Page 25 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

26

Table 2. Calibration test conditions. All tests were conducted at 23±0.5 °C for 72 hours. The 666

“Low Concentration” test was extended to measure continued color change. The “VOC co-667

exposure” test was used to simulate a realistic indoor condition with VOC emissions from a 668

piece of particleboard placed inside the chamber. 669

Test name

Formaldehyde

Concentration (ppb) Co-exposure conditions

Relative

humidity

Target

Actual

(HPLC)

Acetaldehyde

(ppb, HPLC) VOCs

High Concentration 100 109 0 No 50 ±5 %

Medium Concentration 50 68 0 No 50 ±5 %

Low Concentration 25 18 0 No 50 ±5 %

Acetaldehyde co-

exposure 50 68 26 No 50 ±5 %

VOC co-exposure 50 54 18 Yes 50 ±5 %

High relative humidity 50 67 0 No 75 ±5 %

670

671

672

Table 3. Comparison of app layouts before and after refinement. 673

Initial Version Final Version

Two layouts, with many features located on

test page

Three layouts, move surveys to additional

page

Simple list view, only titles and dates Add intuitive color signs and result

information

Three ambiguous steps to take photos Remove capability to take image of

surroundings, highlight the two before/after

steps

Evenly distributed user interface spaces for

all elements

Assign weighted space to each element

Simplified camera process, hard to handle Redesign camera function, show badge icons

674

675

676

677

678

Page 26 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

27

Table 4. Beta test results. A total of 10 people consented to the survey and completed at least 679

one question. 680

Question Yes (n) No (n) No response (n)

Did you have trouble opening the app? 0 10 0

Did you have trouble taking images? 0 9 1

Did you have trouble taking survey? 6 3 1

Was any part confusing?1 2 6 2

Were you able to get a formaldehyde concentration? 8 0 2

Did you receive an error taking survey? 3 1 6

1. Comments indicated confusion was related to survey access or questions. 681

682

FIGURE LEGENDS 683

Figure 1. System Overview. 684

685

Figure 2. Flow within app. Note that the research version also contained a mandatory consent 686

form upon the application start. If the user did not consent, then the application would end. 687

688

Figure 3. Color grid paper used to evaluate lighting conditions. 689

690

Figure 4. The cropped badge image from the SmART-Form application camera. The 691

calibration patch noted in green and the color-changing area is noted in red. Each block 692

contains 50×50 pixels in the 250×250 cropped badge image. 693

694

Figure 5. Final wireframe. 695

696

Figure 6. Screenshots from the app. The first screen is the test display screen. Each test will 697

bring the user to the test detail page (second screenshot). The third screen is accessed via the 698

Page 27 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

28

“data survey” link and will send the user to the external survey after clicking the “go to 699

survey” link. 700

701

Figure 7. Calibration of the app. A. The best-fit line to the data was y = -36301x + 36671 (R² 702

= 0.8811 and P<0.0001) and is shown in small blue dots. Standard deviation lines (dashed 703

outer lines) are shown in green around data. The standard deviation of the data was equivalent 704

to 10.9 ppb at 72 hours of exposure. B. Co-exposure to acetaldehyde or a VOC mixture did 705

not interfere with measurement (P=0.93, P=0.07, respectively). 706

707

Page 28 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

96x79mm (300 x 300 DPI)

Page 29 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

175x359mm (300 x 300 DPI)

Page 30 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

53x35mm (300 x 300 DPI)

Page 31 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

77x70mm (300 x 300 DPI)

Page 32 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

174x291mm (300 x 300 DPI)

Page 33 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

86x51mm (300 x 300 DPI)

Page 34 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

170x214mm (300 x 300 DPI)

Page 35 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

1

Supporting Information for: 1

2

Smartphone-based App for Residential Testing of 3

Formaldehyde (SmART-Form) 4

5

Siyang Zhang1, Nicholas Shapiro

2, Gretchen Gehrke

2, Jessica Castner

3 Zhenlei Liu

4, Beverly 6

Guo4, Romesh Prasad

4, Jianshun Zhang

4, Sarah Haines

5,6,7, David Kormos

6, Paige Frey

8, 7

Rongjun Qin6,9

, Karen C. Dannemiller6,7*

8

9

10 1

Department of Computer Science and Engineering, Ohio State University 11 2

Public Laboratory for Open Technology & Science 12 3

Castner Incorporated 13 4

Department of Mechanical & Aerospace Engineering, Syracuse University 14 5

Environmental Science Graduate Program, Ohio State University 15 6

Department of Civil, Environmental and Geodetic Engineering, College of Engineering, 16

Ohio State University 17 7

Department of Environmental Health Sciences, College of Public Health, Ohio State 18

University 19 8

Strand Associates, Incorporated 20 9

Department of Electrical and Computer Engineering, Ohio State University 21

22

23 *Corresponding author: dannemiller.70@osu.edu, 470 Hitchcock Hall, 2070 Neil Ave, 24

Columbus, OH 43210, 614-292-4031 25

26

27

Page 36 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

2

Contents 28

29

30

Material Page 31

Mathematical Modeling for Color Change Measurement 3 32

Figure S1 3 33

Figure S2 5 34

Figure S3 6 35

Figure S4 6 36

Figure S5 7 37

Figure S6 7 38

Figure S7 8 39

Figure S8 9 40

Figure S9 10 41

Table S1 11 42

Table S2 11 43

44

45

Page 37 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

3

Mathematical Modeling for Color Change Measurement 46

47 In this system, we measure the change of surface albedo, which is calculated as the 48

ratio between the outlet light and inlet light from a surface. The outlet light reflected from the 49

surface into the camera can be referred as the color intensity computed from the Smartphone 50

image. Figure 1 shows the Lambertian reflectance model, a simple and widely used lighting 51

model. In this model, the inlet light and albedo of the material determine the final output light. 52

However, both inlet light ���� and the actual albedo � and �� are unknown and not directly 53

measurable from the images. We will then derive an alternative measurable variable that 54

could be associated with formaldehyde concentration. 55

� ���

��� (1) 56

57

Figure S1. A simple light model 58

From the badge images we can obtain the color intensities (the outlet light) of both the 59

calibration patch and the badge, denoted as �� �� (for calibrating patch) and �� �� (Chemical 60

badge). Let � and � be the albedo of the calibration patch and the chemical badge, we will 61

have the following relationships given Equation (1): 62

� ������

���� , �� �

�����

���� (2) 63

Page 38 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

4

where the index � refers to the ��� image that has a corresponding formaldehyde 64

concentration Ci. Note � is a constant that does not vary with the images, as there will not be 65

any chemical reactions in the calibrating patch. In Equation (2), Since only �� ��� and �� ��

� 66

are known and can be measured from the images, we eliminate the unknowns by taking the 67

first equation in (2) and substitute to the second: 68

�����

����� �

��

� (3) 69

Since � is a constant and �� is directly related to the degree of exposure of badge ��, 70

we then define �� ���

��

�����

����� as the variable (we call it relative albedo) for building the 71

regression model with the formaldehyde concentration ��, and now �� can be computed 72

directly from the images and no actual albedo needs to be explicitly calculated. We tested 73

with the lab samples and build a linear/non-linear regression model. In our modeling process, 74

other variants, such as aperture of the camera, as well as ISO sensitivity settings will also be 75

considered as factors for more accurate modeling. To relax the regression modeling and 76

minimize the compacts from camera variants, we set the camera parameters to fixed value, 77

such as ISO 200, White Balance 5K, Exposure time 1/60s etc. This model will be 78

implemented in the Smartphone app for formaldehyde concentration reading computation. 79

80

Page 39 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

5

81

Figure S2 Diagram of the small environmental chamber system. A formaldehyde permeation 82

tube inside the Dynacalibrator is maintained at a constant temperature with a constant passing 83

flow rate. The formaldehyde-containing airflow from the Dynacalibrator is mixed with the 84

clean air and fed to the 50 L small environmental chamber where the test badges are placed. 85

Different levels of test concentrations are obtained by adjusting the clean airflow rate, and are 86

confirmed by sampling at the chamber exhaust with the DNPH cartridges followed by HPLC 87

analysis (ASTM D5197-16). 88

89

90

91

Page 40 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

6

92

Figure S3 Small-scale environmental chambers (left) for badge exposure and the 93

Dynacalibrator (right) for formaldehyde generation 94

95

Figure S4 Badge sensor inside the environmental chamber 96

97

98

Page 41 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

7

99

Figure S5. Badge readings (color change ratio) read over 64 hours using 25W, 40W, 50W, 100

60W, and 60 W LED lightbulbs. 101

102

103

104

105

Figure S6. Color readings over different camera settings using 25W and 60W. For each image 106

the calibration and chemical path were cropped randomly three times so each of the values in 107

the bar graph are averaged from three values. Error bars (very small) represent the standard 108

error of three randomly selected areas from a single image of each condition. 109 110

Page 42 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

8

111

Figure S7. Screenshots from a previous version of the app. 112

113

114

115

Page 43 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

9

116

Figure S8. Chamber test results for high, medium, and low concentration tests. V1 and V2 117

refer to the two versions of the app (SafeAir®, version 1, and ChromAir®, version 2) that 118

were tested. Images were taken inside the system (through the glass) and then corrected using 119

the images taken outside the glass immediately before/after comparable internal images. 120

121

122

123

Page 44 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

10

124

Figure S9. High relative humidity caused inaccuracies in the readings on both Android and 125

iOS devices. 126

127

128

129

Page 45 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

PROOF

11

Table S1. Boundary values for SmART-Form application variant warning list. Boundaries 130

were selected to be between the extreme values of measurements taken under suitable and 131

unsuitable conditions. 132

Warning Type Selected Boundary Measurements in test for range of

cutoff value

Overexposure Average lightness > 0.8 0.8258-0.6893

Underexposure Average lightness < 0.4 0.3145-0.5520

Badge Contamination

in “Before” image

Lightness ratio > 1 1.0020-0.9651

Blue tint due to high

relative humidity

Average saturation < 0.6 0.5185-0.8008

Short exposure time Exposure time < 12 hours Not applicable

Below detection limit Concentration < 20 ppb Determined by detection limit

Above detection limit Concentration > 120 ppb Determined by tested conditions

133

134

Table S2. Camera settings used along with the color imaging paper in Experiment 3 on an iOS 135

phone. 136 Lighting Experiment Conditions Final Settings

Setting 1 Setting 2 Setting 3 Auto

Adjust

Android <8.0 Android ≥8.0 iOS

White

Balance1

Locked Locked Locked N/A Fluorescent N/A N/A

Color

Temperature2

4003K 4190K 4318K N/A Locked 4000K 4000K

Shutter Speed 1/61s 1/47s 1/400s N/A 1/125 1/125 1/125

ISO3 208 473 555 N/A 200 200 200

Focus Auto Auto Auto N/A Auto Auto Auto

Aperture Lowest Lowest Lowest N/A Lowest Lowest Lowest

137

1 White Balance: Adjusts colors so that the image looks more natural. Most phone cameras 138

have auto, fluorescent, daylight, cloudy, etc. It it similar to the color temperature. 139

2 Color Temperature: Since white balance is locked on iOS systems, we use 4000K color 140

temperature to indicate fluorescent light conditions, similar to typical indoor conditions. 141

3 ISO: the camera’s sensitivity to the light, a higher value represents higher sensitivity. 142

N/A – Not available to access. 143

Page 46 of 46

Indoor Air - PROOF

Indoor Air - PROOF

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960