Diarrheal diseases are a leading cause of under-five mortality and morbidity in sub-Saharan Africa. Quantitative exposure modeling provides opportunities to investigate the relative importance of fecal-oral transmission routes (e.g. hands, water, food) responsible for diarrheal disease. Modeling, however, requires accurate descriptions of individuals’ interactions with the environment (i.e., activity data). Such activity data are largely lacking for people in low-income settings. In the present study, we collected activity data and microbiological sampling data to develop a quantitative microbial exposure model for two female caretakers in peri-urban Tanzania. Activity data were combined with microbiological data of contacted surfaces and fomites (e.g. broom handle, soil, clothing) to develop example exposure profiles describing second-by-second estimates of fecal indicator bacteria (E. coli and enterococci) concentrations on the caretaker’s hands. The study demonstrates the application and utility of video activity data to quantify exposure factors for people in low-income countries and apply these factors to understand fecal contamination exposure pathways. This study provides both a methodological approach for the design and implementation of larger studies, and preliminary data suggesting contacts with dirt and sand may be important mechanisms of hand contamination. Increasing the scale of activity data collection and modeling to investigate individual-level exposure profiles within target populations for specific exposure scenarios would provide opportunities to identify the relative importance of fecal-oral disease transmission routes.
Citation: Julian TR, Pickering AJ (2015) A Pilot Study on Integrating Videography and Environmental Microbial Sampling to Model Fecal Bacterial Exposures in Peri-Urban Tanzania. PLoS ONE 10(8): e0136158. https://doi.org/10.1371/journal.pone.0136158
Editor: Paul J. Planet, Columbia University, UNITED STATES
Received: November 27, 2014; Accepted: July 31, 2015; Published: August 21, 2015
Copyright: © 2015 Julian, Pickering. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Data Availability: All relevant data are within the paper.
Funding: This work was funded by Stanford University’s School of Earth Sciences, Center for African Studies and Woods Institute for the Environment. TRJ received support from the United States Environmental Protection Agency (EPA); EPA has not officially endorsed this publication, and the views expressed herein may not reflect the views of the EPA. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Diarrheal diseases caused by exposure to pathogenic agents are a leading cause of under-five mortality and morbidity in sub-Saharan Africa . In addition, exposure to non-pathogenic fecal bacteria may contribute to child morbidity associated with malnutrition and child stunting . Despite the known and potential impacts of fecal exposures on child health, there is little understanding of the relative importance of fecal-oral disease transmission routes in developing countries . Contemporary research largely focuses on transmission through drinking water and food. However, recent evidence has highlighted the role of non-dietary ingestion pathways (e.g., ingestion of soil and contaminants via hand-to-mouth contacts) in children’s microbial exposures in low income countries [3,4]. In one study, infants in peri-urban areas of Zimbabwe were observed consuming large quantities of fecal bacteria through ingestion of soil and chicken feces . Other studies have demonstrated high concentrations of fecal bacteria on surfaces and soils in developing countries, suggesting these matrices may influence child exposure to fecal contamination [4,6].
Quantitative exposure and risk modeling provides opportunities to identify routes responsible for exposure to feces. Quantitative microbial risk assessment (QMRA) models identify risks associated with infectious diseases exposures . Examples of QMRA models investigating non-dietary ingestion routes include quantifying risks from nosocomial surfaces, food preparation surfaces, and cleaning laundry [8–10]. Risk assessment models have been used to quantify risks associated with specific activities, understand relative contributions of various exposure pathways, highlight the need for quantification of exposure factors, and identify effective interventions [11,12]. For example, Mattioli et al. (2015) applied this framework to estimate that between 97–98% of total fecal matter ingested by a Tanzanian child is due to hand-to-mouth contact events as compared to consumption of contaminated drinking water . One limitation to the accuracy and applicability of risk assessment models is the reliance on accurate descriptions of individuals’ interactions with the environment. Mattioli et al. (2015), for example, modeled Tanzanian children’s interactions with the environment based on data collected about children from the United States . Traditionally, human-environment interaction data are collected by one or more of the following methods: activity recall, activity diaries, structured observation, and third person videography . Videography is considered superior to other activity data collection methods because it is more accurate, eliminates recall bias, and provides an opportunity to record difficult-to-remember events, events of short duration (e.g., hand-to-mouth contacts), and specific sequences of events [14,15]. Third person videography, combined with video translation, has been used to quantify activity data in order to estimate child exposure to chemical contaminants [16,17]. The activity data provided by videography and videotranslation are known as micro-level activity time series data (MLATS). MLATS data are second-by-second sequences of discrete contact events. The high resolution data on sequential contacts of activity data can be applied to understanding dermal and non-dietary exposures. In this study, first person perspective videography (FPV) is employed. FPV shifts the position of the camera from a third person perspective to a first person perspective by using a small, portable, head-mounted videocamera. The videocamera is mounted with a forward and downward angle to capture the full range of motion of the participant’s hands. When correctly framed, the angle of the videocamera is able to capture all hand-to-mouth and object-to-mouth contacts. Although FPV has been previously employed in sociological research, the present study provides the first known instance of using FPV to collect contact data .
The study objective was to demonstrate the application and utility of second-by-second activity data collection to understand fecal contamination exposures on hands in the developing world. The objective is accomplished by piloting FPV to capture activity data to parameterize a quantitative model of fecal bacteria exposures. The study provides a framework for increasing the scale of the described methods (activity data collection and microbial exposure profile modeling) to identify the relative importance of fecal-oral disease transmission routes in low-income countries. Microbial exposure profiles are time series of microbial contamination concentrations on surfaces (for example, hands). The profiles are then used to identify factors influencing contamination and estimate subsequent risks of infection from interactions with the surface. By providing a framework for modeling individual-level exposure profiles, we provide insight into methods needed to better understand inter-individual variability in fecal-oral disease transmission.
Materials And Methods
Two women in a low-income urban community within Dar Es Salaam, Tanzania were recruited coincident with an ongoing field trial of hand hygiene among female caretakers for children under five years old.
The following protocol and consent procedures were approved by both the Stanford University Research Compliance Office for Human Subjects Research and the Muhimbili University’s Institutional Review Board. Verbal informed consent was obtained from both participants following the communication of an approved script in Swahili (the participants’ native language) by a trained field enumerator. Written consent was not necessary because the research fulfilled the Stanford University Human Research Protection Program policy that verbal consent is sufficient when the research presents no more than minimal risk of harm and does not involve any procedures which would otherwise require written consent if conducted outside of the research context. Research proceeded only following affirmation of both understanding and consent, according to the approved protocol.
Video cameras (GoPro Digital Hero 5, Woodman Labs, San Mateo, CA) were attached to head mounts such that the camera angled forward and downward to capture the majority of the hands’ range of motion (Fig 1). The two participants were instructed to complete their normal activities for two consecutive thirty-minute intervals from approximately 8:30 AM until 9:30 AM. In total, two hours of video were collected.
A modified version of Virtual Timing Device (VTD) Software (SamaSama Consulting, Sunnyvale, CA) was used to translate the videotape footage into MLATS for the caretakers’ right and left hands . The VTD Software provides an interface consisting of two grids; each grid contains cells that represent various options for a) location of female caretaker, and 2) objects contacted by hand of caretaker. Each location and object is represented by a button on the software interface and can be specified by the user. The two locations in the first grid included: 1) inside and 2) outside. The second grid specified objects, which included surfaces that a hand contacted, water that a hand was immersed in (e.g., when washing clothing), and other special designations (e.g., the hand is not touching anything).
The objects that were specified for this study were chosen during an initial screening of the videos based on what objects were most frequently contacted. The eighteen surfaces in the objects grid included: 1) wooden broom, 2) metal bucket, 3) burlap sack, 4) charcoal, 5) clothing, 6) dirt or sand, 7) wooden door, 8) participant’s own face (face), 9) food scraps, 10) participant’s own hands, 11) money, 12) plastic objects, 13) metal cooking utensils including cooking pots and handles, 14) rubber, 15) bar soap, 16) stone, 17) wooden objects, and 18) paper towel used for hand sampling. The four water immersion objects included: 1) washing clothing, 2) washing hands, 3) water for drinking, and 4) water for hand sampling. The two special designations included: 1) nothing, and 2) not in view. To use VTD Software, a researcher views the video of the caretaker and in real time selects buttons for the location and object the participant is contacting. VTD records the sequence of the objects contacted along with the duration that the object is contacted. Contacts are recorded for all events where fingers, palm, and/or back of the hand come into direct contact with an object. Only one object can be contacted at a time, and when no object is contacted, the object category “nothing” is selected. The category “not in view” is selected when the hand is out of view and it is unclear what object the hand is contacting. Output from VTD is a computer text file with each line providing the location of the caretaker, the object contacted, and the duration of the contact. The present study provides data on the objects contacted by the left and right hands of the female caretakers. Data from VTD were imported into Microsoft Excel 2007 for analysis. As is common with video-translation using VTD, quality control measures are required (Beamer 2012). There was one researcher who translated the two hours of videotape. The same researcher then reviewed the tapes second-by-second to ensure the translations were accurate. When errors were identified, short segments (15 seconds to 2 minutes) were translated a second time, integrated into the data file to replace the corrected segment, and the file was reviewed again.
Hand samples were obtained from participants prior to collection of video. The hand samples were obtained using the hand rinse sampling method, which includes contacts with both water (characterized in the exposure model as the object ‘water used for hand sampling’) and a paper towel (characterized as the object ‘paper towel used for hand sampling’). . In brief, the hand rinse method includes placing hands consecutively in a Whirl-pak bag (Nasco, Fort Atkinson, WI, USA) containing 350 ml of Uhaibottled drinking water pre-screened for the absence of target bacteria (i.e., E. coli and enterococci) and dosed with sodium thiosulfate. E. coli and enterococci were chosen as target bacteria because they are indicative of fecal contamination and co-occur with fecal pathogens on hands [19,20]. Additionally, both E. coli and enterococci are more abundant and easier to quantify than fecal pathogens. Both hands were placed into one bag, as opposed to measuring contamination on each hand individually, to decrease the lower limit of detection for bacterial contamination.
In addition to hand rinses, five fomites were sampled (see Table 1). Samples were chosen from fomites that participants had contacted during the video recording, as observed by the data collectors. The samples were collected to provide data on bacterial contamination of fomites contacted by the participants. Therefore, the samples were chosen as a representative subset of all fomites that participants had contacted during the video recording, as observed by the data collectors. Fomite samples were obtained by swabbing a surface of roughly 10 cm x 10 cm with polyester-tipped swabs pre-wet in 10 ml of ¼-strength Ringer’s solution (Oxoid Limited, Hampshire, UK) . Hand rinse and fomite samples were transported on ice to the laboratory for analysis within 4 hours of sample collection.
All samples were processed via membrane filtration to enumerate enterococci using mEI media per EPA method 1600  and E. coli using MI media per EPA Method 1604  as described elsewhere [3,19]. Lower and upper limits of detection for E. coli and enterococci on fomites are estimated at approximately 2.5 and 500 CFU/100 cm2. Limits of detection for bacteria on fomites were calculated using a countable range of 1–200 colonies following filtration of 2/5ths (2 ml out of 5 ml) of Ringer’s solution used for fomite sampling. Lower and upper limits of detection for bacteria on hands are estimated at 3.5 and 700 CFU / 2 hands. Limits of detection on hands were calculated using a countable range of 1–200 colonies following filtration of 2/7ths (100 ml out of 350 ml) of water used for hand sampling.
Exposure Assessment Model
The exposure model is a modified version of the conceptual model for estimating microbial exposures from fomites previously described in Julian et al. (2009). Two exposure models were developed separately, one model for E. coli and one for enterococci contamination of hands. In brief, every object identified using videotranslation was assigned a quantitative estimate of surface E. coli and enterococci concentration and a quantitative estimate of the fraction of bacteria transferred on contact (Tables 1 and 2). Using the micro-level activity time series data, we assume the sequential contact of each object transferred bacteria to or from the hands based on the surface concentrations of the object and hand and the object-specific transfer efficiency consistent with the work of Julian et al. (2009) . Similarly, we assumed parameters besides surface type, transfer direction, and magnitude did not influence transfer (e.g., wetness) . We assumed every contact impacted the concentration of bacteria on the hand: bacteria transferred from hand-to-object when bacterial concentrations were greater on hands than objects and from object-to-hand when concentrations were greater on objects than hands . The amount of bacteria transferred to or from the hand was determined by multiplying the concentration gradient between the hand and object by the transfer efficiency :
Where CHf is the final concentration on the hand (CFU/100cm2), CHi is the initial concentration on the hand (CFU / 100cm2), CO is the concentration on the object (CFU/100cm2), and T is the transfer efficiency for the object (%). The assumption of transfer from the surface with greater contamination to that with lesser contamination is based on previous modeling work as there are limited data on the impact of existing surface contamination on the magnitude or direction of transfer between two surfaces composed of different materials. The time series of E. coli and enterococci concentrations on the surfaces of the each hand, as determined by both sequential contact events and bacterial inactivation, is the output from the exposure model . Bacteria concentrations on hands are then converted, using an estimate of sampling efficiency, to concentrations measured via the hand sampling method (see Surface Concentration).
Surface bacteria concentrations for each object category were estimated. The ten fomites that were sampled (Table 1 “Measured”) were used to estimate surface concentrations for eleven object categories (wooden broom, metal bucket, burlap sap, clothing, dirt or sand, wooden door, food scraps, plastic objects, metal cooking utensils, rubber, and wooden objects). Because E. coli recovery from surfaces using cotton-tipped swabs removes only an estimated 20% of bacteria , the measured concentrations were divided by a sampling efficiency to estimate the actual surface contamination (Table 1 “Modeled”), using:
Where FI is the initial concentration of the fomite, FM is the concentration of bacteria measured on the fomite, and fF is the sampling efficiency for E. coli using swabs (Table 3). Concentrations of bacteria for five categories (not in view, nothing, paper towel used for hand sampling, bar soap, and water used for hand sampling) were set equal to 0 CFU/100 cm2 or 0 CFU/100 ml under the assumption the surfaces were clean. Notably, even when bar soap is contaminated with bacteria, there is no detectable bacteria transfer to hands on contact [25,26]. Concentrations of bacteria for three categories (washing clothes, washing hands, and water used for drinking) were set equal to 25 CFU/100ml and 200 CFU/100ml for E. coli and enterococci, respectively, based on mean concentrations of E. coli and fecal streptococci, a group which includes enterococci, in stored water in Tanzania . The object category for ‘their own hands’ was assigned a concentration equal to the concentration predicted by the exposure model for the other hand at the time of contact. Concentrations for the remaining four categories (charcoal, their own face, money, and stone), for which we have no data, were assumed to be contaminated at the lower detection limit (2.5 CFU/100 cm2), corresponding to contamination of 12.5 CFU/100 cm2 when accounting for sampling efficiency (Table 1).
Transfer efficiencies of bacteria between hands and objects were based on a literature review of transfer efficiencies in high relative humidity (relative humidity in Dar es Salaam, Tanzania in July 2009 was between 60% and 90%, Table 2) [19,20,27–29]. The transfer rate of bacteria between hands and water when the hands are immersed in contaminated water is based on the findings of O’Toole et al. (2009) . Transfer efficiency for hand washing is based on the findings of Pickering et al. (2010) and Pickering et al. (2011) for female caretakers in Tanzania [19,20]. Although transfer efficiency of bacteria between hands and surfaces has been shown to be influenced by the type of surfaces, inoculum size, and characteristics of the contact event (pressure, friction, wetness), we simplified the model by neglecting these characteristics in line with our previous work which suggested that model output is relatively insensitive to transfer efficiency magnitude when multiple contacts are modeled [23,30].
The model assumed that bacterial contamination of hands is impacted by contact, with bacteria transferring from the surface (object or hand) with the higher concentration of bacteria to the surface (object or hand) with the lower concentration. However, only the concentrations of bacteria on hands were assumed to be impacted; the concentrations of objects were assumed to remain at the measured concentration. For hands, the initial concentrations were calculated based on the results of the hand sampling at time 0, and were adjusted based on hand sampling removal efficiencies of 52% for E. coli and 44% for enterococci . Based on estimated removal efficiencies, initial hand concentrations were calculated using:
Where CI is the initial concentration on hands (CFU/2 hands), CM is the measured concentration on hands using the hand sampling method (CFU/2 hands), and fH is the hand sampling removal efficiency. Concentrations are then adjusted to units of colony forming units (CFU) per cm2 based on average hand surface areas of 440 cm2 .
We also accounted for inactivation on hands consistent with the model previously reported by Julian et al. (2009) . The log10 inactivation rate (kEC) for E. coli was assumed to be 3 x 10−3 1/s, based on the findings of Pinfold (1990), where 99% of E. coli were inactivated on hands in 10 minutes . A lower log10 inactivation rate (kENT) for enterococci of 1.7 x 10−4 1/s was assumed based on a 50% decrease in fecal streptococci in 30 minutes also reported in Pinfold (1990) .
To account for surface area during contact events, we assumed that a contact event transferred only the portion of bacterial contamination on the object in direct contact with the hand. For object contacts, we assumed an average 44 cm2 of contact area (10% of 440 cm2, the total hand surface area). For water contacts, we assumed total hand submersion (440 cm2, or100% of total hand surface area) [33,34]. We assumed duration of contact did not influence the fraction of bacteria transferred, consistent with the work of Cohen-Hubal et al. who found duration did not impact chemical residue transfer on contact . This assumption was extended to include immersion of hand with water: we assumed duration of immersion did not impact fraction of bacteria transferred. However, immediately after the transfer event, we assumed that the bacteria transferred were uniformly distributed over the hand, in line with the previous model 
Non-dietary ingestion of fecal bacteria was modeled assuming that all hand-to-own face contact events resulted in ingestion of bacteria from hands. The amount of bacteria ingested was modeled based on previous work in Julian et al. (2009):
Where D is the dose of bacteria ingested, SH is the surface area of the hand in contact with the mouth, TEHF is the percentage of bacteria transferred from the hand to the mouth, and CH is the concentration of bacteria on the hands. Similar to transfer of bacteria between surfaces, we again assume the surface area of hand-to-mouth contact events equals 10% of the total hand surface area. Other sources of fecal bacteria ingestion (such as mouth contacts with contaminated objects besides hands, ingestion of contaminated water, and ingestion of contaminated food) were not included in the model because these events did not occur during the observation.
Micro-level Activity Time Series
Participant A was recorded for 54:16 min and spent her time sweeping and washing laundry (Table 4). Participant B was recorded for 63:16 min and spent her time making metallic oil lamps for resale (Table 4). The left hand for Participant A contacted 16 different object categories 200 times. The right hand for Participant A contacted 18 different object categories 208 times. The left and right hands for Participant B contacted 11 and 15 different objects, respectively, 271 and 368 times.
Both hands of both participants spent 10–20% of the total observed time not in contact with any objects or surfaces (see Table 4). Specifically, the left and right hands of Participant A were observed to spend 19% (10:23) and 12% (6:44), respectively, of the total observed time not contacting any object. Similarly, the left and right hands of Participant B were observed to spend 16% (10:08) and 11% (7:02), respectively, not contacting any object.
For the majority of time, both hands for both participants were in full view of the camera (see Table 4). However, full extension of the arm and misplacement of the camera on the head during the second half of the video for one participant (Participant B) resulted in the hands frequently falling outside of the camera’s field-of-view. Participant A’s left and right hands were not in view 3 times each for a total duration of 0:13 (0.4%) and 0:20 (0.6%), respectively. Conversely, Participant B’s left and right hands were not in view 52 and 31 times for a total duration of 8:26 (13%) and 4:30 (7.1%), respectively.
Enterococci and E. coli on Hands and Fomites
The initial enterococci concentration on Participant B’s hands was 340CFU / 2 hands. The corresponding E. coli concentration was above the limit of detection (>700 CFU / 2 hands) and therefore assumed to be 700 CFU / 2 hands. For Participant A, both initial enterococci and E. coli concentration were above the detection limit (>700 CFU / 2 hands) and so were also assumed to be 700 CFU / 2 hands. Both E. coli and enterococci concentrations on fomites ranged from <2 to >500 CFU/100 cm2 (see Table 1).
Measured concentrations on both hands and fomites were adjusted for the exposure model to account for sampling efficiency (Table 3). Initial concentrations for bacteria on hands ranged from 773 to 1591 CFU per 2 hands, or 387 to 796 CFU per hand. Fomite concentrations used for the exposure model, after accounting for sampling efficiency, ranged between 0 and 2500 CFU/100cm2.
The modeled concentrations on the hands of each participant for E. coli and enterococci follow similar temporal trends. The concentration of bacteria increases substantially after 10 to 15 minutes, coincident with the participants’ contacts with dirt/sand (Fig 2). The bacteria concentrations then decrease. The decrease is more rapid for Participant B than for Participant A. The decreases are either due to contacts with surfaces with lower bacterial concentrations than on the hands (Participant A) or contacts with water (Participant B). The bacteria concentrations eventually reach levels near the detection limit and then remain low for both participants for the remainder of the study.
Fecal bacteria ingested is estimated based on hand-to-own face contact events. Participant A contacted her own face 4 times corresponding to an estimated cumulative ingestion of 46 CFU E. coli (ingestion events of 23, 22, 0.6, and 0.6 CFU E. coli) and 65 CFU enterococci (ingestion events of 33, 31, 0.6, and 0.7 CFU E. coli). Participant B contacted her own face 6 times corresponding to cumulative ingestion of 148 CFU E. coli (ingestion events of 24, 39, 18, 49, 16, and 3 CFU E. coli) and 38 CFU enterococci (ingestion events of 8, 14, 5, 8, 2, and 1 CFU enterococci).
Exposure to fecal contamination in the developing world is linked to poor child and maternal health [1,2]. In the present study, first person perspective videography is used to collect micro-level activity time series (MLATS) data and MLATS data are used to develop exposure models. The study highlights evidence that exposure to dirt or sand may be most responsible for contamination of caretaker’s hands and that water-related activities (e.g., washing clothing, washing hands) may not be effective at dramatically reducing E. coli contamination on hands. The impact of washing clothing, in particular, on E. coli concentrations may be due to E. coli contamination of laundry as observed in the United States and/or by use of soil or sand as a cleaning aid, as observed in Peru [36,37]. These findings further support the work of Pickering et al. (2011) that shows female caretakers’ activities can dramatically influence their hand contamination .
First person videography is capable of capturing MLATS for hands of study participants. When the camera is appropriately placed on the participant’s forehead, as was the case for Participant A, first person videography is able to capture more than 99% of all hand contacts. However, FPV is subject to more data loss than third person perspective videography when the camera is misplaced, as was the case for Participant B. Misalignment of the camera for Participant B impacted the entire hour duration resulting in substantial data loss: 13% of the total time for the left hand and 7% for the right hand. As a comparison, the time out of view for third person perspective videography for child exposure factors is approximately 9% (range of 2 to 18%) for the left hand and 7% (2–13%) for the right hand . Two modifications would improve first person videography as a tool for collecting MLATS: a camera with a wider field of view (wide angle lens) and a headset that adjusts to the individual participant and reduces risk of misalignment. Two other concerns for FPV are distractions associated with wearing a camera on one’s head and invasions of personal privacy. Although neither participant appeared, nor described themselves as, distracted or concerned about privacy, a larger sample size of participants would be needed to understand these concerns. Future studies should include adaptation of protocols to allow for temporary removal of video cameras as well as survey-based data collection from participants on issues concerning FPV surveillance compliance.
Given the small sample size of two participants, generalizable conclusions drawn from the collected MLATS are limited. The potential range of activities for female caretakers in low-income countries is large, incorporating not only the activities observed but other domestic (e.g., shopping, cooking, water collection) and/or economically-productive (e.g., agricultural, commercial) activities. Nevertheless, similarities between the participants were observed, including high levels of activity characterized by frequent repeat contacts of common objects (e.g., metal bucket objects, metal cooking utensils, brooms, clothes) and infrequent contacts with rare objects (e.g., charcoal, dirt/sand, money, stone). This is noteworthy as the exposure models suggest that infrequent contacts (e.g, contact with dirt and sand in Fig 2, Participant B) were responsible for dramatic increases in bacterial contamination on hands. The reliance on a single video translation observer may have influenced categorization of objects contacted which were not validated via inter-observer, but accuracy of the frequency and duration of contact events was ensured via the described intra-observer validation protocol.
Estimates of parameter values strongly influence model outputs. The data used in the model were drawn from a combination of literature review, microbiological sampling, and videography. Of these sources, the microbiological sampling data are likely both the most influential and least certain. First, we did not collect estimates of microbial surface contamination for all of the fomites that the female caretakers contacted (e.g., charcoal, money, stone, faces). In the future, a wider array of surfaces should be sampled to ensure all surfaces contacted have an accurate estimate of microbial contamination. Second, bacteria concentrations on hands exceeded the upper limit of detection for the assays. We therefore had to assume that bacteria concentration on hands was equal to the upper limit of detection for the exposure models. This assumption directly influenced the performance of the model as our assumption that E. coli hand contamination was at the upper detection limit may have been an underestimate. Underestimating initial E. coli concentrations likely resulted in a model that systematically underestimated hand contamination for the remaining time series. In the future, assays should be designed to increase the upper limit of detection of hand contamination so that an accurate estimate of initial hand contamination can be used to initialize and investigate the accuracy of the model.
There are notable limitations to our work. One limitation is that we measured bacterial contamination on hands from both hands simultaneously instead of measuring bacterial contamination of each hand separately. Because the left and right hands are not necessarily equally contaminated, the initial concentrations for each hand may be biased. Having separate data points for each hand individually would have enabled comparison of modeled to estimated hand contamination for each hand. A second limitation is that our model is not validated. Although we attempted to validate the model by collecting hand samples at the middle and end of the videography (data not shown), the data suffered from limited interpretability. This was because only two additional samples were collected for each exposure profile, the samples represented contamination on both hands, and half of the measured concentrations were outside of the countable range. Future studies looking to validate exposure models should collect hand samples at a timescale relevant to expected changes in microbial contamination. Our model, for example, suggests dirt/sand contacts dramatically increase bacterial contamination.
Exposure models provide insight into the relative importance of contact events on microbial contamination of hands. The work presented here highlights the disproportional role on infectious disease exposures that infrequent contact events may have. Given the small sample size of this study, it is important to consider increasing the scale of activity data and microbial data collection. Applying the model to develop individual-level exposure profiles within target populations would provide opportunities to identify human-environment interactions that most influence fecal-oral disease transmission routes.
The authors acknowledge the volunteers, as well as Sara Marks and Marlene Wolfe for help with data collection.
Conceived and designed the experiments: TRJ AJP. Performed the experiments: TRJ AJP. Analyzed the data: TRJ. Contributed reagents/materials/analysis tools: TRJ AJP. Wrote the paper: TRJ AJP.
- 1. Mathers CD, Boerma T, Ma Fat D (2008) The Global Burden of Disease: 2004 Update. Geneva, Switzerland: World Health Organization.
- 2. Humphrey JH (2009) Child undernutrition, tropical enteropathy, toilets, and handwashing. Lancet 374: 1032–1035. pmid:19766883
- 3. Mattioli MCM, Pickering AJ, Gilsdorf R, Davis J, Boehm AB (2012) Hands and water as vectors of diarrheal pathogens in Bagamoyo, Tanzania. Environ Sci Technol 47: 355–363.
- 4. Pickering AJ, Julian TR, Marks SJ, Mattioli MC, Boehm AB, Schwab KJ, et al. (2012) Fecal contamination and diarrheal pathogens on surfaces and soil show spatial heterogeneity within Tanzanian households and no association with improved sanitation. Environ Sci Technol 46: 5736–5743. pmid:22545817
- 5. Ngure F (2012) Environmental Hygiene, Food Safety And Growth In Less Than Five Year Old Children In Zimbabwe And Ethiopia: Cornell University.
- 6. Sinclair RG, Gerba CP (2010) Microbial contamination in kitchens and bathrooms of rural Cambodian village households. Lett Appl Microbiol 52: 144–149. pmid:21198693
- 7. Haas CN, Rose JB, Gerba CP (1999) Quantitative Microbial Risk Assessment. New York, NY: John Wiley and Sons, Inc.
- 8. Gibson LL, Rose JB, Haas CN (1999) Use of quantitative microbial risk assessment for evaluation of the benefits of laundry sanitation. Am J Infect Control 27: S34–S39. pmid:10586144
- 9. Chen Y, Jackson KM, Chea FP, Schaffner DW (2001) Quantification and variability analysis of bacterial cross-contamination rates in common food service tasks. J Food Protect 64: 72–80.
- 10. Nicas M, Sun G (2006) An integrated model of infection risk in a health-care environment. Risk Anal 26: 1085–1096. pmid:16948699
- 11. Jones RM, Adida E (2012) Selecting nonpharmaceutical interventions for influenza. Risk Anal.
- 12. Nicas M, Jones RM (2009) Relative contributions of four exposure pathways to influenza infection risk. Risk Anal 29: 1292–1303. pmid:19558389
- 13. Mattioli MCM, Mattioli J, Davis A, Boehm (2015) Hand-to-Mouth Contacts Result in Greater Ingestion of Feces than Dietary Water Consumption in Tanzania: A Quantitative Fecal Exposure Assessment Model. Environmental science & technology 49: 1912–1920.
- 14. Zartarian VG, Streicker J, Rivera A, Cornejo CS, Molina S, Valadez OF, et al. (1995) A pilot-study to collect micro-activity data of 2-year-old to 4-year-old farm-labor children in Salinas Valley, California. J Expo Anal Env Epid 5: 21–34.
- 15. Beamer P, Key ME, Ferguson AC, Canales RA, Auyeung W, Leckie JO (2008) Quantified activity pattern data from 6 to 27-month-old farmworker children for use in exposure assessment. Environ Res 108: 239–246. pmid:18723168
- 16. Shalat SL, Stambler AA, Wang Z, Mainelis G, Emoekpere OH, Hernandez M, et al. (2011) Development and in-home testing of the pretoddler inhalable particulate environmental robotic (PIPER Mk IV) sampler. Environ Sci Technol 45: 2945–2950. pmid:21351766
- 17. Beamer PI, Canales RA, Bradman A, Leckie JO (2009) Farmworker children's residential non-dietary exposure estimates from micro-level activity time series. Environ Int 35: 1202–1209. pmid:19744713
- 18. Fathi A, Hodgins JK, Rehg JM. Social interactions: A first-person perspective; 2012. IEEE. pp. 1226–1233.
- 19. Pickering AJ, Davis J, Walters SP, Horak HM, Keymer DP, Mushi D, et al. (2010) Hands, water, and health: Fecal contamination in Tanzanian communities with improved, non-networked water supplies. Environ Sci Technol 44: 3267–3272. pmid:20222746
- 20. Pickering AJ, Julian TR, Mamuya S, Boehm AB, Davis J (2011) Bacterial hand contamination among Tanzanian mothers varies temporally and following household activities. Trop Med Int Health 16: 233–239. pmid:21091858
- 21. Agency UEP (2002) EPA method 1600: Enterococci in water by membrane filtration using membrane-enterococcus indoxyl-β-D-glucoside agar (mEI). Washington, D.C.: U.S. Environmental Protection Agency. EPA 821-R-02-022 EPA 821-R-02-022.
- 22. Agency UEP (2002) EPA method 1604: Total coliforms and Escherichia coli in water by membrane filtration using simultaneous detection technique (MI medium). Washington, D.C.: U.S. Environmental Protection Agency. EPA 821-R-02-024 EPA 821-R-02-024.
- 23. Julian TR, Canales RA, Leckie JO, Boehm AB (2009) A model of exposure to rotavirus from nondietary ingestion iterated by simulated intermittent contacts. Risk Anal 29: 617–632. pmid:19187484
- 24. Moore G, Griffith C (2007) Problems associated with traditional hygiene swabbing: the need for in‐house standardization. J Appl Microbiol 103: 1090–1103. pmid:17897214
- 25. Feachem RG (1984) Interventions for the control of diarrhoeal diseases among young children: promotion of personal and domestic hygiene. Bull World Health Organ 62: 467. pmid:6331908
- 26. Heinze JE, Yackovich F (1988) Washing with contaminated bar soap is unlikely to transfer bacteria. Epidemiol Infect 101: 135–142. pmid:3402545
- 27. O'Toole J, Sinclair M, Leder K (2009) Transfer rates of enteric microorganisms in recycled water during machine clothes washing. Appl Environ Microbiol 75: 1256–1263. pmid:19124592
- 28. Gerardo U. Lopez CPG, Akrum H. Tamimi , Kitajima Masaaki, . Maxwell Sheri L, and Rose Joan B. (2013) Transfer efficiency of bacteria and viruses from porous and nonporous fomites to fingers under different relative humidity. Appl Environ Microbiol 79: 5728–5734. pmid:23851098
- 29. Rusin P, Maxwell S, Gerba C (2002) Comparative surface-to-hand and fingertip-to-mouth transfer efficiency of Gram-positive bacteria, Gram-negative bacteria, and phage. J Appl Microbiol 93: 585–592. pmid:12234341
- 30. Pérez-Rodríguez F, Valero A, Carrasco E, García RM, Zurera G (2008) Understanding and modelling bacterial transfer to foods: a review. Trends in Food Science & Technology 19: 131–144.
- 31. Agency UEP (2011) Exposure Factors Handbook. US Environmental Protection Agency Washington, DC.
- 32. Pinfold JV (1990) Faecal contamination of water and fingertip-rinses as a method for evaluating the effect of low-cost water supply and sanitation activities on faeco-oral disease transmission. I. A case study in rural north-east Thailand. Epidemiol Infect 105: 363–375. pmid:2209740
- 33. Agency UEP (2006) Child-Specific Exposure Factors Handbook Washington, D.C.: U.S. Environmental Protection Agency EPA 600-R-06-096A EPA 600-R-06-096A.
- 34. AuYeung W (2007) Analysis and Simulation of Young Children's Hand and Mouthing Contact Behaviors. [Ph. D. Thesis]. Stanford, CA: Stanford University.
- 35. Cohen Hubal EA, Nishioka MG, Ivancic WA, Morara M, Egeghy PP (2008) Comparing surface residue transfer efficiencies to hands using polar and nonpolar fluorescent tracers. Environ Sci Technol 42: 934–939. pmid:18323125
- 36. Cardoso CC, Fiorini JE, Ferriera LR, Gurjao J, Amaral LA (2000) Disinfection of Hospital Laundry Using Ozone: Microbiological Evaluation•. Infect Control Hosp Epidemiol 21: 248–248.
- 37. Julian TR, MacDonald LH, Guo Y, Marks SJ, Kosek M, Yori PP, et al. (2013) Fecal indicator bacteria contamination of fomites and household demand for surface disinfection products: A case study from Peru. Am J Trop Med Hyg 5: 869–872.
- 38. Ferguson AC, Canales RA, Beamer P, AuYeung W, Key M, Munninghoff A, et al. (2006) Video methods in the quantification of children's exposures. J Expo Sci Env Epid 16: 287–298.