Risk factors in heating, ventilating, and air-conditioning systems for occupant symptoms in US office buildings: the US EPA BASE study


Mark J. Mendell
Lawrence Berkeley National Laboratory
1 Cyclotron Rd.
MS 90-3058 Berkeley
CA 94720
Tel.: 1-510-486-5762
Fax: 1-510-486-6658
e-mail: mjmendell@lbl.gov


Episodes of symptom complaints, including upper and lower respiratory symptoms, eye and skin irritation, headache, and fatigue, have been reported for decades by occupants of office buildings in many countries. Explaining and mitigating these problems have been challenging. Numerous scientific studies have documented that these building-related symptoms (BRS), sometimes called sick building syndrome, are surprisingly common even in buildings without widespread health complaints (Burge et al., 1987). Specific measured indoor exposures causing these symptoms have not been scientifically documented, in either ‘complaint’ buildings or buildings in general. Researchers have, however, identified a number of environmental ‘risk factors’ that are correlated with higher prevalence of symptoms in buildings selected without regard to occupant complaints, and thus may indicate as yet unidentified causal exposures. These risk factors include the presence of air-conditioning systems (Seppanen and Fisk, 2002), contaminated components of heating, ventilating, and air-conditioning (HVAC) systems (Mendell et al., 2003; Sieber et al., 1996), low ventilation rates (Seppanen et al., 1999), higher temperatures (Mendell et al., 2002a,b), and dampness/visible mold in buildings (Mendell et al., 2006; Park et al., 2004).

Presence of central mechanical air-conditioning systems in office buildings (relative to natural ventilation) is one of the risk factors associated most consistently with increased BRS (Mendell and Smith, 1990; Seppanen and Fisk, 2002), although substantial variation in BRS exists among buildings with air-conditioning. Current causal hypotheses focus on aspects of HVAC that may influence production, dissemination, or concentration of indoor biological and chemical contaminants. Regarding these hypotheses, much epidemiologic evidence is available related to low ventilation rates as risk factors (Seppanen et al., 1999) and limited evidence is available related to moisture in HVAC systems (Mendell et al., 2006) and poor condition and maintenance of HVAC systems (Mendell et al., 2003).

The US Environmental Protection Agency’s Building Assessment and Survey Evaluation (BASE) study, the largest study to date of building environments and occupant symptoms in a representative set of US office buildings, offers a unique opportunity to investigate risk factors for BRS. This paper presents findings from an analysis in the BASE data of risk factors involving features or practices in buildings that (i) could be determined by inspection or interview and (ii) were hypothesized to be related to contamination or performance of HVAC systems. These factors include various aspects of design, condition, maintenance, and operation of HVAC systems. We hypothesized that risk factors likely to increase the probability of biological or chemical contaminants disseminated by HVAC systems, or to decrease effectiveness of ventilation that would remove indoor contaminants, would be correlated with irritant, allergic, or toxic responses in occupants. These would thus correlate with increased prevalence of work-related symptoms, potentially including lower and upper respiratory symptoms, cough, eye symptoms, fatigue or difficulty concentrating, headache, and skin symptoms. Because many of these risk factors would be correlated with each other, and because many non-environmental factors in office buildings are known to be associated with symptom reporting, we used multivariate models to estimate the independent associations with BRS of the HVAC-related risk factors.

While several HVAC factors related to moisture and related contamination have been associated with increased symptoms in prior analyses of these or other data (e.g. poorly draining condensate drain pans, residue in drain pans, dirt and contamination at outdoor air intakes, infrequently cleaned coils and pans) (Mendell et al., 2003, 2006; Sieber et al., 1996), predicted associations of symptoms with most other aspects of HVAC systems investigated here were conjectural, based on engineering judgment. Thus, the analyses reported here were primarily to explore previously untested hypotheses.


The BASE study data were collected from 1994 to 1998 by the US EPA in 100 large US office buildings. Descriptions of this study have been published previously (Brightman and Moss, 2000) and full details are available in the study protocol (US Environmental Protection Agency 2003). Briefly, the study selected a representative set of 100 office buildings from geographic regions throughout the USA, and randomly selected within each building a study space with, if possible, at least 50 occupants and served by no more than two ventilation air-handling units. The terms ‘building’ and ‘study space’ in this analysis are thus essentially equivalent and used interchangeably. Data were collected from questionnaires given to all occupants of each study space, from standardized inspections of the buildings and ventilation systems, from standardized interviews conducted with facility managers and from a broad range of environmental measurements.


Analyses used outcomes of ‘weekly, building-related symptoms’– defined as specific symptoms experienced in the building at least 1 day per week in the last 4 weeks and also improving away from the building. Analyses reported here investigated associations of risk factors with seven building-related symptom outcomes, representing either single or combined symptoms from the questionnaire: lower respiratory (at least one of the three symptoms wheeze, shortness of breath, or chest tightness); cough; upper respiratory (at least one of the three symptoms stuffy or runny nose, sneezing, or sore or dry throat); dry, itching or irritated eyes; fatigue or difficulty concentrating; headache; and dry, itching, or irritated skin.

Risk factors

Analyses investigated potential risk factors related to design, operation, maintenance, and condition of HVAC systems. Data on these risk factors were collected by study personnel from inspection of ventilation systems, buildings, and occupied spaces, or through interviews with facility managers on building and ventilation system-related practices and history. Table 1 lists the HVAC variables in the BASE data of initial interest and indicates how each was handled in the process of defining risk factor variables for analysis. Because of missing, inconsistent or illogical data values, insufficient variation, or strong intercorrelations, original BASE variables for many risk factors were either excluded from analyses or combined in composite variables or indices. Where possible, compound variables and indices corresponding to concepts of hypothetical interest were created. Both original and newly created variables with continuous values were generally analyzed as categorical variables.

Table 1.   HVAC system risk factors initially considered – handling in the analyses
How variable handled in analysesHVAC design or configurationHVAC condition, maintenance, or operation
Potential source of contaminantsPotential influence on ventilationPotential source of contaminantsPotential influence on ventilation
Excluded before bivariate models because of missing data, inadequate variation, or inconsistent information (thus not evaluated for associations with symptoms)Type of ductwork in central air handler
Return duct material
Building exhaust systemFrequency of cleaning for all specific HVAC components (except cooling coils, drip pans, humidifiers, and cooling towers)Frequency of testing and balancing
Minimum outdoor air intake rate
Correct fan direction
Morning purge cycle
Omitted after bivariate analyses. from all modelsNearby pollution sourcesVariable-air-volume vs. constant-air-volume  
Included in initial risk selection analyses as individual variablesSupply duct material
Outdoor air intake strategy
HVAC configuration (e.g. roof-top units, through-the-wall units, etc.)
Outdoor flow control strategy (economizers)Condition of: filtration system; cooling towers; air handler/duct liner
Frequency of: filter replacement cooling tower cleaning
Filter fit
Average daily hours of ventilation
Frequency of controls calibration
Included in initial risk selection analyses as combined or grouped variablesPresence and type of local coils and drain pans
Height of outdoor air intake location
Humidification type
Floor area per operable windowCondition of air-handling unit components and ducts
Frequency of: scheduled inspection of HVAC components; cleaning of coils/condensate drain pans
Condition/maintenance/cleaning frequency of humidification systems, if present
Operational condition of HVAC components

For instance, we created a variable not in the original data set for height above the ground of the outdoor air intake, by combining available data items on location of the intake (ground, wall, or roof) and height of the intake above the adjacent ground or rooftop, and then adjusting initial values based on field notes, drawings, and photographs. Where a study space was supplied by multiple air handlers with different intake heights, we averaged those heights. The continuous values of this variable were represented as a four-category variable in analyses, using indicator variables to represent distances above the ground ranging from 3 m below to more than 60 m above. To assess any risks or benefits associated with operable (i.e. openable) windows, to go beyond a simple count or a proportion of windows that were operable, we combined available variables for number of operable windows in each study space and occupied floor area per study space, to create a variable for the area of occupied space per operable window. This variable was ultimately dichotomized (essentially as operable vs. non-operable windows), with a reference category including 32 study spaces with floor areas of 0–600 m2 per operable window. The other category combined one study space with almost no operable windows (i.e. floor area of 1175 m2 per operable window) with the 64 buildings with no operable windows.

A single analysis variable created for humidification systems combined information from four original BASE variables to group buildings in the following three categories: buildings with humidifiers that had the worst combination of poor condition, infrequent inspection, and infrequent cleaning; buildings with fair or good combinations of these humidifier qualities; and buildings with no humidification. Two composite variables were created related to the visible condition of HVAC components: the cleanliness/condition of HVAC components, as an indicator of risk of contamination of the ventilation air from dirty or wet surfaces, and the operational condition of HVAC components, as an indicator of risk from inadequate operation and potentially reduced provision of ventilation air.

The BASE data set includes outdoor concentrations of selected pollutants, including volatile organic compounds (VOCs), measured for each study space at the approximate location and height of its outdoor air intake for each study space, as well as indoor concentrations at three randomly selected locations within each study space. The VOC measurements were based on analyses of time-integrated samples collected during an approximate 10-h period on Wednesday of the study week (US Environmental Protection Agency 2003). Two sampling methods were used: SUMMA canisters and (only in 70 buildings) multisorbent tubes.

Confounding variables

We considered several directly measured environmental parameters as potential confounding variables in analyses: temperature (using a variable for average number of hours*degrees per day that the indoor temperature was above 20°C) (Apte et al., 2000), humidity ratio (a measure of absolute humidity uncorrelated with temperature, calculated from measured temperature and relative humidity, based on mean indoor levels on one day), and outdoor air ventilation rate (VR) [using a variable for ventilation estimated from measures of volumetric flow (Mendell and Lei, 2005)]. Additional potential environmental confounders included season of study and age of building. Personal variables from the occupant questionnaire on demographics, health status, job, and workspace factors were also considered in analyses as potential confounding variables: gender, age, education, smoking status, asthma, mold allergy, hay fever, type of workstation, comfort of chair, satisfaction with work station, job satisfaction, job demand, job conflict, and years worked in building. Several personal variables were also considered as potential effect modifiers in the modeling process: gender, smoking, asthma, allergy, and years worked in building. Outdoor air ventilation rate was included among potential confounding variables, although it may have been in the causal pathway for at least four variables (variable-air-volume vs. constant-air-volume system, average daily hours of ventilation, operational condition of HVAC components, proportion of operable windows). To check this, we also planned to run a set of parallel models without adjustment for ventilation rate to see if the estimated associations for those risk factors changed with VR in the model.

Analysis methods

Details of analytical procedures are provided in Appendix 1. In summary, we first investigated and selected or combined potential risk factors using univariate analyses and bivariate analyses with all symptom outcomes. We then further selected individual risk factors, separately for each symptom outcome, in bivariate logistic regression models, and then reduced the sets of risk factors for each outcome in separate multivariate logistic regression (‘risk selection’) models. We created the final logistic regression models for each symptom outcome model by consideration of potential interaction terms and the selection and addition of confounding variables. These models were then rerun as logistic regression/general estimating equation (GEE) models. A set of alternate models without ventilation rate as a confounding variable was also planned. Estimates from final logistic regression/GEE models are reported in tables and text as odds ratios (ORs) and 95% confidence intervals (CIs). OR values exceeding 1.0 indicate increased symptom prevalence in subjects with the risk factor; values less than 1.0, a decreased prevalence; and values = 1.0, no relationship.

Finally, in exploring some associations identified in the analyses between symptoms and outdoor air intake height, we plotted several sets of data: the crude prevalence of upper respiratory symptoms by the outdoor air intake height, and values of several volatile organic compounds typically emitted by motor vehicles (e.g. benzene, toluene, ethylbenzene, and o-, p-, and m-isomers of xylene, collectively referred to as BTEX), measured both inside each study space and outside each building (at about the height and location of the outdoor air intake for the study space).


BASE data were available from 4326 building occupants and on study spaces within 100 buildings, although the three naturally ventilated study spaces without HVAC systems were excluded from these analyses. (Although differences in symptom prevalence between naturally ventilated and air-conditioned buildings are of great interest, there were too few naturally ventilated buildings available to make meaningful comparisons.) The overall response rate for the occupant questionnaire was 85%. Overall prevalence (in the entire survey population) of the seven symptom outcome definitions investigated ranged from a low of 4.2% for lower respiratory symptoms to a high of 20.9% for upper respiratory symptoms (see bottom row of Table 2). Prevalence of each outcome varied substantially among the 100 individual buildings.

Table 2.   Unadjusted odds ratios (ORs) and 95% confidence intervals (CIs) for associations between HVAC risk factors and occupant symptoms in US office buildings in the BASE data, collected 1994–1998
Potential risk factorsNumber of buildings/ study spacesWeekly, building-related symptom outcomes
Lower respiratoryCoughUpper respiratoryIrritated or itching eyesFatigue or difficulty concentratingHeadacheIrritated or itching skin
OR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CI
  1. *P-value < 0.05.

  2. aLocal cooling coils refer to the presence of any type of system with cooling coils in the study space, with or without a drain pan, including those described as air–water induction units, fan coil units, individual room packaged AC units, heat pumps, or other unitary systems.

  3. bConditioned means that the outdoor air is filtered and possibly dehumidified, heated, or cooled before delivery to the air handler, whereas unconditioned denotes lack of these features. Intake fan denotes a separate fan to bring in the required amount of outdoor air; with no intake fan, outdoor air is drawn in by suction induced by the supply fan.

  4. cThe ‘central humidification system’ variable combined values of the following variables: air handler has humidification system, general condition of humidifiers, frequency of humidifier inspection, and frequency of humidifier cleaning.

  5. dThe ‘cleanliness/condition of HVAC components’ variable combined values of the following variables: general condition of mechanical room, general condition of air handler housing, condition of air handler coils, general condition of air handler components, condition of air handler drain pans, condition of air handler intakes, general condition of air distribution ductwork, and condition of humidifier drain pans.

  6. eThe ‘operational condition of HVAC components’ variable combined values of the following variables: condition of air handler dampers, condition of air handler fan belts, leakage at seams of air distribution ductwork, condition of exhaust fan belts, general condition of exhaust fans, particle filtration system change label, general condition of terminal units, condition of terminal unit dampers, general condition of control systems, and condition of control system sensors.

  7. fThe ‘frequency of scheduled inspection of HVAC components’ variable combined values of variables for inspection frequency of: air handler housing, heating/cooling coil, drain pan, air distribution ductwork, control systems, cooling tower, fan coil unit, and terminal units.

HVAC design
Height of outdoor air intake above ground
 >60 m121.00 1.00 1.00 1.00 1.00 1.00 1.00 
 >30 to <=60 m152.34*1.23–4.451.560.86–2.842.42*1.72–3.411.47*1.05–2.061.76*1.22–2.531.280.88–1.872.25*1.11–4.54
 >0 to <=30 m601.310.73–2.351.420.85–2.382.19*1.62–2.961.57*1.18–2.091.67*1.22–2.271.58*1.16–2.162.11*1.13–3.94
 >−3 to <=0 m92.81*1.46–5.401.540.82–2.892.43*1.70–3.481.50*1.05–2.151.86*1.28–2.721.82*1.24–2.662.51*1.23–5.15
Local cooling coilsa
 No local coils 101.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Local coils871.050.63–1.741.260.76–2.101.77*1.33–2.381.50*1.12––1.481.49*1.08–2.050.860.55–1.36
Supply duct material
 No flexible duct261.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Flexible only610.910.65–1.270.850.62–1.150.82*0.69–0.970.880.74–1.060.960.79–1.150.940.77–1.140.980.71–1.35
 Flexible and fiberboard 100.710.40–1.290.570.32–1.010.67*0.50–0.890.71*0.52–0.960.67*0.48–0.920.940.69–1.290.48*0.25–0.93
Outdoor air intake strategyb
 Unconditioned and with intake fan71.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Conditioned and with intake fan80.470.20–1.120.640.30–1.350.59*0.40–0.870.840.54–1.301.150.72–1.821.050.66–1.690.570.25–1.30
 Unconditioned and with no intake fan770.890.50–1.590.880.51–1.520.850.63––1.681.350.93–1.961.280.88–1.861.010.57–1.81
Floor area per operable window
 <600 m2/operable window321.00 1.00 1.00 1.00 1.00 1.00 1.00 
 >=600 m2/operable window or none651.000.73–1.370.800.60–1.070.77*0.66–0.890.85*0.72–1.000.82*0.69–0.971.030.86––1.52
Intake control of outdoor air
 Economizer681.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Fixed minimum intake231.110.79–1.560.840.60–1.180.960.81–1.150.870.72–1.050.900.74–1.090.870.71–1.060.980.71–1.37
 100% outdoor air or other60.770.35–1.670.520.23–1.180.59*0.40–0.880.770.52–1.130.62*0.40–0.960.830.55–1.240.570.25–1.30
Air handler system type
 CAV291.00 1.00 1.00 1.00 1.00 1.00 1.00 
 VAV or mixed680.850.61–1.160.840.62––1.220.920.77–––1.331.210.87–1.67
HVAC condition, maintenance, or operation
Central humidification systemc
 No humidification system821.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Humidified, maintenance/condition good or fair 90.610.32–1.180.920.54–1.560.990.75–1.290.840.62–1.130.840.62–1.151.331.00–1.760.640.35–1.20
 Humidified, maintenance/condition poor 61.280.80–2.052.06*1.40–3.012.05*1.63–2.571.87*1.48–2.381.88*1.47–2.411.74*1.34–2.251.80*1.20–2.69
Cleanliness/condition of air handler componentsd (average for components)
 Good to fair391.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Fair to poor271.290.92–1.821.180.85–1.641.000.84–1.200.950.78–1.140.860.70–1.050.880.72–1.090.940.66–1.34
Liner condition in air handler housing and duct
 Good401.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Fair or poor521.160.85–1.591.220.91–1.641.23*1.05–1.431.080.92–––1.311.140.84–1.53
Operational condition of air-handling system components based on inspectione (average for components)
 Good to fair591.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Fair to poor151.300.87–1.951.080.73–1.610.920.74––1.250.860.67–1.100.780.60––1.87
Cooling towers: general condition
 None present301.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Fair or poor351.000.69–1.440.980.71–1.360.81*0.68–0.970.860.72–1.050.930.76–1.130.900.73–1.110.940.66–1.34
Particulate filtration system: general condition
 Good531.00 1.00 1.00 1.00 1.00 1.00 1.00 
Particulate Filtration System: Filter Fit
 Good711.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Fair or poor261.40*1.02–1.941.000.73–1.380.960.81–––1.210.920.76––1.46
Frequency of cleaning of cooling coils and condensate drain pans
 Semi-annually or more101.00 1.00 1.00 1.00 1.00 1.00 1.00 
 As needed or none472.03*1.01–4.061.510.86–2.671.110.85–1.451.160.87–1.550.960.72––1.621.100.64–1.86
Frequency of cleaning of cooling tower
 No cooling tower301.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Semi-annually or more201.030.67–1.571.010.69–1.480.910.74–1.120.920.74––1.311.080.85–1.371.200.81–1.77
 As needed or none101.020.58–1.760.800.47–1.370.68*0.51–0.910.940.71–1.260.840.61–1.150.67*0.47–0.960.530.27–1.05
Frequency of scheduled inspection of HVAC componentsf (average for components)
 Semi-quarterly or more361.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Quarterly to annually 471.120.80–1.580.780.56–1.070.950.80–1.120.970.81–1.160.860.72–1.030.910.75–1.100.980.71–1.35
 As needed or none141.61*1.06–2.431.71*1.20–2.451.69*1.37–2.082.00*1.61–2.481.230.98–1.551.150.90–1.471.270.85–1.90
Frequency of control system calibration
 Semi-annually or more141.00 1.00 1.00 1.00 1.00 1.00 1.00 
 As needed or none710.950.63–1.430.990.67–1.471.110.90–1.370.880.71–1.090.980.78–1.230.980.77––1.58
Frequency of replacement of panel filters
 Semi-quarterly or more121.00 1.00 1.00 1.00 1.00 1.00 1.00 
 Quarterly or semi-annually630.830.55–1.250.820.55–1.210.840.68–1.050.800.64–1.010.890.70–1.130.930.72–1.190.790.53–1.18
 Annually or less160.570.32–1.020.850.52–1.410.68*0.51–0.900.65*0.48–0.870.840.62–1.140.790.57–1.090.870.52–1.44
Average hours of ventilation operation on weekdays
 24 h311.00 1.00 1.00 1.00 1.00 1.00 1.00 
 13–20 h350.67*0.47–0.980.900.65–1.250.920.77–1.090.81*0.67–0.980.860.71–1.050.910.74–1.110.850.60–1.20
 10–12 h310.980.69–1.390.870.62–1.230.900.75–1.080.870.72–1.060.980.80––1.230.960.68–1.36
Outcome prevalence in overall study population 4.2% 5.0% 20.9% 18.3% 16.4% 14.9% 4.6% 

The study population has been described elsewhere in more detail (Brightman and Moss, 2000). To summarize, 66% of respondents were female, 61% were between the ages of 30 and 49, and 15% were current smokers. Respondents had worked in their buildings for an average of 5.9 years. Respondents’ job categories were: 35% professionals, 34% clerical, 17% managers, and 14% technical. Regarding highest level of education, approximately 46% of respondents had less than a college degree, 36% had an undergraduate degree, and 18% had a graduate degree.

We organized the risk factors into two categories: HVAC design or configuration, and HVAC condition, maintenance, or operation. Table 1 lists the risk factors of initial interest, represented by 70 variables in the BASE data and the way each was ultimately handled in the analysis: either by exclusion before the bivariate analyses for any of a number of reasons, exclusion after bivariate analyses for lack of association with any outcome or inclusion in at least one risk selection model as individual variables or within a combined or grouped variable.

Descriptive analysis

Table 2 shows the number of buildings/study spaces in each category of the risk factor variables. All variable categories contained six or more buildings. Numbers for each variable may not add to 97 because of missing values. Although the three study spaces ventilated only by operable windows were excluded, 33 of the 97 study spaces included had some operable windows, even though all were air-conditioned. Eighty-seven study spaces had some kind of local cooling coils (these are distributed at multiple locations throughout a building, rather than located only in central air handlers in mechanical rooms, on roofs, etc.) Fifteen study spaces had central humidification systems, of which six (40%) were at the lowest level on a combined index of inspection frequency, cleaning frequency, and condition. Eighty-six spaces had cooling coils or drain pans cleaned only annually or without a regular schedule. Outdoor air intakes for the study spaces were mostly 0–30 m above ground level, but nine study spaces had intakes at or below ground level, and 27 study spaces had intakes more than 30 m above ground level, of which 12 were above 60 m.

Unadjusted analyses

Table 2 shows unadjusted (bivariate) associations with outcomes for the risk factor variables selected or created after univariate analyses. Among HVAC design or configuration factors, outdoor air intake height categories less than 60 m above the ground were strongly associated with most symptom outcomes, with some suggestion of the highest risks for intakes below ground level. Study spaces with local cooling coils were associated with increased upper respiratory symptoms, eye symptoms, and headache. Compared with supply ducts containing no flexible duct, those including only flexible material were associated with some decreased risk of upper respiratory symptoms, and the combination of flexible and fiberboard duct was associated with a decrease in at least four symptom outcomes. In buildings with no (or almost no) operable windows, prevalence of upper respiratory symptoms, eye symptoms, and fatigue/difficulty concentrating decreased slightly.

Among HVAC condition variables, the presence of a humidification system in good condition, relative to absence of humidification, was associated with some increase in headache. Humidification systems in poor condition, however, were associated with substantial increase in most symptoms. Average cleanliness/condition of air handler components rated as fair relative to good-to-fair was associated with a small unexpected reduction in four symptom outcomes; however, systems in the poorest condition were not consistently associated with changes in symptoms.

Among HVAC maintenance factors, cleaning of cooling coils and drain pans scheduled only annually was associated with increase in multiple symptoms, although the absence of regularly scheduled cleaning was associated with increase only in lower respiratory symptoms. Lack of regularly scheduled inspection of HVAC system components was associated with substantial increases in upper and lower respiratory symptoms, cough, and eye symptoms. Less frequent replacement of panel filters was associated with at least some reduced prevalence for all symptoms, especially with annual or less frequent replacement.

For each symptom outcome model, bivariate model outcomes determined which variables were included in initial risk selection models (not shown), using criteria described in Appendix 1. All risk factors then retained in each risk selection model were kept throughout the construction of final models for that outcome.

Adjusted analyses

Table 3 shows estimates from final multivariate logistic regression/GEE models, along with the numbers of individuals and buildings (=study spaces) included in each final model. Each final symptom outcome model described in Table 3 contains only the risk factors for which ORs are provided in the table, along with confounding variables selected for that model; these are not included in the table. Ventilation rate, when treated as a potential confounding variable, was not selected as a confounder for inclusion in any final model. Thus, the question of whether adjustment for ventilation rate would be problematic because it was in the causal pathway for several risk factors became moot. Hosmer–Lemeshow goodness-of-fit P-values for final models ranged from 0.34 to 0.97, without requiring alteration of models produced by the basic algorithm. Many fewer associations were evident after multivariate adjustment than in the bivariate models. We found no meaningful effect modification for selected risk variables and the personal variables for gender, smoking, asthma, allergy, and years in building.

Table 3.   Adjusted odds ratios (ORs) and 95% confidence intervals (CIs), from logistic regression models with generalized estimating equationsa, for associations between HVAC risk factors and occupant symptoms in US office buildings in the BASE data, collected 1994–1998
Risk factorWeekly, building-related symptom outcomes
Lower respiratoryCoughUpper respiratoryIrritated or itching eyesFatigue or difficulty concentratingHeadacheIrritated or itching skin
OR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CIOR95% CI
  1. *P-value < 0.05.

  2. aFor each of the seven symptom outcomes, only potential risk factors with estimates provide in this table are included in the model, along with confounding variables selected during model construction. Potential confounding variables (not all included in each model) include environmental variables (temperature, humidity ratio, ventilation rate, season of study, and age of building) and personal variables (gender, age, education, smoking status, asthma, mold allergy, hay fever, type of workstation, comfort of chair, satisfaction with work station, job satisfaction, job demand, job conflict, and years worked in building). Ventilation rate was not selected for inclusion in any outcome model.

  3. bSee Table 2, footnote a.

  4. cSee Table 2, footnote b.

  5. dSee Table 2, footnote c.

  6. eSee Table 2, footnote d.

  7. fSee Table 2, footnote e.

  8. gSee Table 2, footnote f.

HVAC design
Height of outdoor air intake above ground
 >60 m1.00   1.00   1.00 1.00 1.00 
 >30 to <=60 m2.04*1.01, 4.11  2.67*1.75, 4.091.64*1.07, 2.531.74*1.18, 2.561.320.88, 1.982.47*1.11, 5.47
 >0 to <=30 m1.140.59, 2.21  1.98*1.37, 2.861.400.96, 2.041.74*1.23, 2.471.69*1.18, 2.412.62*1.22, 5.59
 >−3 to <=0 m1.870.91, 3.86  2.11*1.40,, 1.812.13*1.41, 3.211.74*1.15, 2.622.250.97, 5.22
Local cooling coilsb
 No local cooling coils     1.00     1.00   
 Local cooling coils     1.370.94, 2.01    1.380.98, 1.93  
Supply duct material
 No flexible duct    1.00         
 Flexible only    0.880.70, 1.11        
 Flexible and fiberboard    1.060.74, 1.51        
Outdoor air intake strategyc
 Unconditioned and with intake fan      1.00     1.00 
 Conditioned and with intake fan      1.460.78, 2.75    0.750.25, 2.24
 Unconditioned and with no intake fan      1.721.00, 2.98    0.940.45, 1.95
Floor area per operable window              
 <600 m2/operable window            1.00 
 >=600 m2/operable window or none            1.85*1.24, 2.76
HVAC condition, maintenance, or operation
Central humidification systemd
 No humidification system  1.00 1.00 1.00 1.00 1.00 1.00 
 Humidified, maintenance/condition good to fair   1.090.61, 1.941.140.84, 1.550.910.64, 1.290.790.56, 1.101.340.98, 1.850.47*0.23, 0.96
 Humidified, maintenance/condition poor   1.310.79, 2.171.49*1.09, 2.051.46*1.05, 2.031.67*1.23, 2.261.340.97, 1.851.65*1.02, 2.68
Cleanliness/condition of air handler components (average for components)e
 Good to fair1.00 1.00 1.00 1.00   1.00   
 Fair0.810.53, 1.230.840.56, 1.270.75*0.59, 0.960.74*0.59, 0.93  0.74*0.58, 0.94  
  Fair to poor1.110.75, 1.651.160.75, 1.781.010.78, 1.320.870.67, 1.13  0.950.72, 1.25  
Liner condition in air handler housing and duct              
 Good  1.00 1.00         
 Fair or poor  1.150.81, 1.651.43*1.13, 1.83        
Operational condition of air-handling system components based on inspection (average for components)f
 Good to fair    1.00     1.00   
 Fair    1.130.91, 1.41    1.200.95, 1.51  
 Fair to poor    0.70*0.49, 0.98    0.760.54, 1.09  
Frequency of cleaning of cooling coils and condensate drain pans
 Semi-annually or more1.00     1.00 1.00 1.00   
 Annually1.920.89, 4.15    1.67*1.15, 2.411.360.97, 1.911.57*1.09, 2.25  
 As needed or none1.560.71, 3.42    1.48*1.01,, 1.591.47*1.03, 2.11  
Frequency of cleaning of cooling tower              
 No cooling tower            1.00 
 Semi-annually or more            1.290.81, 2.05
 Annually            1.060.70, 1.62
 As needed or none            0.530.25, 1.10
Frequency of scheduled inspection of HVAC components (average for components)g
 Semi-quarterly or more1.00 1.00 1.00 1.00 1.00     
 Quarterly to annually 1.170.80, 1.700.750.52, 1.070.940.76,, 1.360.75*0.61, 0.91    
 As needed or none1.500.95, 2.371.480.98, 2.241.60*1.21, 2.102.18*1.55, 3.060.880.67, 1.16    
Frequency of replacement of panel filters
 Semi-quarterly or more    1.00 1.00       
 Quarterly or semi-annually    1.020.76, 1.361.180.89, 1.56      
 Annually or less    1.070.72, 1.590.840.59, 1.20      
Average weekday hours of ventilation operation
 24 h1.00       1.00 1.00 1.00 
 13–20 h0.800.53, 1.20    0.900.71, 1.150.880.70, 1.090.930.74, 1.170.670.45, 1.02
 10–12 h1.180.79, 1.77    1.120.87, 1.451.32*1.05, 1.661.210.94, 1.541.190.77, 1.84
Number of observations (occupants) in final model4,0223,9873,7183,6884,1064,0183,884
Number of buildings in final model  95  92  85  84  95  95  95

Among HVAC design or configuration factors, outdoor air intake heights less than 60 m above the ground were generally associated with substantially increased odds for multiple symptom outcomes, including fatigue/concentration difficulty (ORs 1.7–2.1), upper respiratory symptoms (ORs 2.0–2.7), skin symptoms (ORs 2.2–2.6), and lower respiratory symptoms (ORs 1.1–2.0). For distances less than 60 m, risks did not show systematic monotonic increases with intakes closer to the ground. HVAC systems with local cooling coils were associated with a suggestion of increased headache [OR (CI) = 1.38 (0.98–1.93)] and upper respiratory symptoms [OR (CI) = 1.37 (0.94–2.01)]. Unconditioned intake of outdoor air with no intake fan was associated with an increase in eye symptoms [OR (CI) = 1.72 (1.00–2.98)]. In buildings with no or very few operable windows, prevalence of skin symptoms [OR (CI) = 1.85 (1.24–2.76)] increased. Other HVAC design factors did not show clear adjusted associations with symptom prevalence.

Among HVAC condition variables, the presence of a humidification system with good condition/maintenance, relative to no humidification, was associated with a significant decrease in skin symptoms [OR (CI) = 0.47 (0.23–0.96)], and a suggestion of an increase in headache [OR (CI) = 1.34 (0.98–1.85)]. Humidification systems with poorer condition/maintenance, however, were associated with some increase in most symptoms, especially fatigue/difficulty concentrating [OR (CI) = 1.67 (1.23–2.26)], skin symptoms [OR (CI) = 1.65 (1.02–2.68)], upper respiratory symptoms [OR (CI) = 1.49 (1.09–2.05)], and eye symptoms [OR (CI) = 1.46 (1.05–2.03)]. Among other HVAC condition factors, cleanliness/condition of air handler components was not consistently associated with symptoms, although fair condition relative to good-to-fair condition was associated with a small reduction in headache, upper respiratory symptoms, and eye symptoms. Having lining in the ducts and air handlers in only fair or poor condition was associated with some increase in upper respiratory symptoms [OR (CI) = 1.43 (1.13–1.83)]. Although there was a suggestion of an increased risk for HVAC components in fair operational condition relative to the best condition category (good-to-fair), fair-to-poor operational condition was associated with some unexpected decrease in upper respiratory symptoms [OR (CI) = 0.70 (0.49–0.98)].

Among HVAC maintenance factors, cleaning of cooling coils and drain pans scheduled only annually or not scheduled, relative to at least semi-annually, were associated with increase in headache [OR (CI) = 1.57 (1.09–2.25) and 1.47 (1.03–2.11)], and eye symptoms [OR (CI) = 1.67 (1.15–2.41) and 1.48 (1.01–2.17)], respectively, and also with some apparent increase in lower respiratory symptoms and, possibly, fatigue/difficulty concentrating symptoms. Inspection of HVAC system components scheduled only quarterly to annually relative to more often was not associated with increased symptoms except for a small reduction in fatigue/concentration difficulty: OR(CI) = 0.75 (0.61–0.91); however, lack of a regular maintenance schedule was associated with substantial increases in eye symptoms [OR (CI) = 2.18 (1.55–3.06)] and upper respiratory symptoms [OR (CI) = 1.60 (1.21–2.10)], along with possible increases in lower respiratory symptoms and cough. Operation of the ventilation system for between 13 and 20 h per day, relative to 24-h operation, was associated with slightly lower prevalence of several symptoms, although confidence intervals were broad. Operation for only 10–12 h per day, however, was associated with slightly higher prevalence of several symptoms, especially fatigue/difficulty concentrating [OR (CI) = 1.32 (1.05–1.66)] and possibly headache.

Other results

Figure 1 plots the crude prevalence of upper respiratory symptoms as a function of outdoor air intake height. Symptom prevalence was substantially higher at intake heights less than about 60 m above ground level and reasonably homogeneous within that range. This relationship also generally held for other symptoms we investigated (not shown). Figures 2–4 show data from 96 buildings with HVAC systems and available data on indoor and outdoor VOC concentrations from SUMMA canister measurements. Figure 2 shows a plot of the average concentrations of o-xylene, measured outside each building (at about the height and location of the outdoor air intake for the study space), as a function of outdoor air intake height. The concentrations are reasonably homogeneous at heights up to about 60–65 m above the ground level and then decrease at higher levels. Results for the other vehicle-related compounds investigated were generally similar. Figure 3, which plots the same data as Figure 2 but using indoor concentrations of o-xylene, shows a similar pattern within a higher range of concentrations. Figure 4 shows indoor concentrations of o-xylene plotted against outdoor concentrations, with a crude trend line suggesting that they are positively correlated. The outdoor and indoor concentrations, however, each measured in two different ways, did not show consistent positive correlations with crude prevalence of symptoms reported by occupants in the study spaces (not shown). Measurements based on SUMMA canister samples showed a positive relationship with upper respiratory symptoms for most BTEX compounds except benzene; however, measurements based on multisorbent tube samples in 70 buildings did not show such relationships.

Figure 1.

 Plot of crude prevalence of building-related upper respiratory symptoms as a function of height above ground level of the outdoor intake, in 100 US office buildings in the US EPA BASE Study

Figure 2.

 Plot of average outdoor concentration of o-xylene (sampled with SUMMA canisters) as a function of height above ground level of the outdoor intake, in 96 mechanically ventilated US office buildings in the US EPA BASE Study

Figure 3.

 Plot of average indoor concentration of o-xylene (sampled with SUMMA canisters) as a function of height above ground level of the outdoor intake, in 96 mechanically ventilated US office buildings in the US EPA BASE Study

Figure 4.

 Plot with estimated linear trend line for average indoor concentration of o-xylene as a function of average outdoor concentration (sampled with SUMMA canisters), in 96 mechanically ventilated US office buildings in the US EPA BASE Study


Comparison with prior findings and predictions

The EPA BASE data allow the first broad assessment in US office buildings of the associations between suspected indoor environmental risk factors and non-specific symptoms in office workers. The present analysis primarily investigated aspects of HVAC systems not previously assessed for associations with symptoms and selected a large number of potential risk factors through engineering judgment of what HVAC characteristics might plausibly increase indoor exposures to contaminants [ventilation rates in BASE, for instance, have been the subject of separate analyses (Apte et al., 2000; Mendell and Lei, 2005)]. Thus, the primary goal of this analysis was to explore a large number of preliminary hypotheses, and simultaneous inclusion in models of all the potential risk factors was not possible. Construction of these models thus did not involve forcing all hypothesized risk factors into models, but selection based on relationships in the data.

A few risk factors included in the present analyses have been investigated previously. Prior analyses of the BASE study data, using slightly different definitions for symptom outcomes and risk factor categories, found similarly elevated risks for less frequently cleaned condensate drain pans and coils (Mendell et al., 2006). Prior analyses of other data, from ‘complaint’ office buildings investigated for health problems reported by occupants, found elevated lower respiratory symptoms associated with several aspects of poor HVAC maintenance, including poorly draining condensate drain pans and debris in air intakes (Mendell et al., 2003). Similarly increased odds ratios for these risk factors were not found in the present analyses, in initial crude models with specific HVAC condition variables; these variables were then combined into composite variables in the current analyses. The higher contamination levels found in ‘complaint’ buildings may be unusual in ‘normal’ buildings and thus not detectable in a relatively small study of 100 buildings.

Prior studies have found presence of humidification systems to be associated with increased symptoms (Mendell and Smith, 1990), although the relationship has not been consistent (Seppanen and Fisk, 2002). Analyses here examined the risks of humidification systems, including the quality of system maintenance and condition. To our knowledge this has not been reported before. Poorly maintained humidification systems were associated with increases in most symptoms assessed, while well-maintained humidification systems had only a marginal adverse association with increased headache. Humidification is considered to reduce dryness-related symptoms in cold-winter climates. The only apparent protective association of well-maintained humidifiers, however, was with skin symptoms. If these findings were to be replicated, consideration of humidification systems would need to balance this potential benefit (53% reduction in odds of skin symptoms, although with a 34% increase for headache) seen with the well-maintained humidifiers in this study, against the apparently larger potential adverse effects associated with the poorly maintained humidification systems (31–65% increase in odds for six symptoms, including skin, seen in almost half of the buildings with humidification systems).

Local cooling coils have been suspected to increase risks because of the difficulty of cleaning and maintaining large numbers of wet coils distributed throughout a building, often in hard-to-access locations such as in ceilings. Only one study has reported investigating the association between local cooling units (e.g. fan coil units) in office buildings and symptoms (Burge et al., 1987). That study found this to be one of several types of HVAC configuration associated with the elevation of a ‘total symptom index’, relative to naturally ventilated buildings. The current analysis found local cooling units were associated only with some possible increase in headache and upper respiratory symptoms, although the reference was other air-conditioned buildings, not naturally ventilated buildings, which could be expected to have generally lower symptom prevalence (Seppanen and Fisk, 2002).

Lack of operable windows was associated only with increased skin symptoms. Despite widespread anecdotal dissatisfaction of occupants with sealed windows in office buildings, there have been almost no prior epidemiologic studies of symptom prevalence related to sealed windows. This may be because sealed windows and air-conditioning, another risk factor for increased symptoms in offices, are too closely correlated in practice to disentangle statisticially. Zweers et al. (1992) in a multivariate analysis of data from Dutch office buildings of different ventilation types, also found that sealed windows were associated with increases in skin symptoms. It is not clear what mechanism would explain this. Other risk factors for which our findings accord roughly with anecdotal beliefs among building investigators include HVAC systems lacking regularly scheduled inspections (increased eye symptoms, upper and lower respiratory symptoms, and cough), and poorer condition (e.g. dirty or wet) of lining in ducts and air handler housings (increased upper respiratory symptoms).

The most striking and consistent finding, which to our knowledge has not been previously investigated or reported, was that outdoor air intakes less than 60 m above the ground level were associated with, on average, approximately 40–140% increased odds for all symptom outcomes. Risks did not increase consistently as intake heights were closer to or below the ground level (60 m down to −3 m). It is not apparent what could explain this strong effect, seen with air intakes even eight to 15 stories (30–60 m) above the ground. This consistent association may result from intake of outdoor air pollutants, such as vehicle-related pollutants, that are at higher concentrations near ground level in cities. Ground-level vehicular exhaust pollutants, to explain the current findings, would need to be evenly mixed well above 30 m above the ground level but at lesser concentrations higher up. Outdoor measurements across the different BASE buildings were consistent with such an outdoor gradient (as in Figure 2), although they did not include measurements at different heights outside the same buildings. A study by Rubino et al. (1998) provides data relevant to this question. This study monitored concentrations of vehicle-related pollutants in outdoor air, for 30 days, at various heights outside a 100-m tall urban building. They reported decreasing concentrations of automotive-related pollutants with increasing height. As distance above ground level increased from 0 to 80 m, airborne concentrations of PM10 decreased steadily from 40 to 32 μg/m3, but for distances between 80 and 110 m, PM10 dropped suddenly to about 25 μg/m3. BTEX compounds were approximately 25% lower (97 vs. 136 μg/m3) at heights of 75–105 m than at 15–45 m. CO concentrations decreased about 29% as height increased from 4 to 104 m above ground level, with the steepest decrease up to 20 m, but some continued reduction up to about 50 m.

The data in Figures 1–4 do suggest that outdoor vehicle-related VOCs, when taken in through low outdoor air intakes, may increase indoor exposures to these pollutants. The data also suggest that these VOCs or other correlated vehicle-emitted pollutants such as CO or particles, may have a role in the occurrence of building-related symptoms. Exposures to the compounds in vehicular emissions, however, have to our knowledge not yet been previously shown to cause increased symptoms at the low indoor concentrations (i.e. 2–4 ppb) associated with these increased outdoor concentrations. The indoor concentrations of o-xylene in the BASE buildings (range mostly 0–3 ppb) are often but not always higher (Figure 4) than the corresponding outdoor concentrations near the air intakes (mostly 0–1.5 ppb), presumably because of indoor sources. A similar trend was observed for the other BTEX compounds measured at the BASE buildings. The associations found here of outdoor air intakes below 60 m and increased occupant symptoms need replication and explanation.


The BASE data, although the largest and most comprehensive collection of standardized data on indoor environments and occupants in representative office buildings in the USA, has many limitations for helping understand causes of building-related health effects. The BASE study was conducted primarily to obtain normative data rather than to test specific a priori hypotheses. Thus many building features of research interest cannot be investigated because they do not occur commonly enough in the study buildings. BASE data were collected on many questions, so that each often lacks sufficient detail or accuracy to answer specific hypothetical questions. There are also inherent difficulties in studying environments as complex as large buildings: Although the BASE study contained over 4000 individual occupants in the study spaces, available resources allowed collection and analysis of most environmental data only at the level of the 100 buildings or study spaces rather than at the level of individual workspaces, allowing limited analysis of variation in these factors. The environmental reports from inspection are subjective and imprecise, and the resulting inaccuracies could have resulted in bias toward the null, obscuring true associations. This is also true of the subjective, self-reported health outcome assessments used.

With respect to limitations in the analyses, the many intercorrelated environmental factors assessed could often not be included in the same models, making it impossible to assess risks for some factors of interest while holding other closely related factors constant. Finally, this analysis included many risk factors, leading to the possibility of false positive associations occurring by chance alone (Rothman, 1990). If the number of risk factors investigated in the analyses are taken to be the 37 terms for potential risk factors included in initial bivariate models for seven outcomes, then if no true underlying associations existed and if all estimates were independent, chance alone would predict approximately 13 associations with P < 0.05 (i.e. 37*7/20) in final models. Thus, of the 32 associations with P-values < 0.05 in the final multivariate models, about 40% may have been false positives. We would suggest that the associations least likely to be false positives are those reported previously from other data or found for multiple symptom outcomes.


This analysis explored various aspects of HVAC systems as potential risk factors to explain the consistent previous finding that the presence of HVAC systems was associated with, on average, increased building-related symptoms among occupants. Nevertheless, HVAC systems are considered necessary in buildings to maintain thermal comfort, control humidity, and provide outdoor air to maintain good indoor environmental quality. Prior available evidence suggests either production of contaminants (Mendell et al., 2003) or less effective ventilation, by some HVAC systems, as the most likely explanation. However, little epidemiologic investigation of this question has occurred. The present analysis suggests several aspects of HVAC systems (poorly maintained humidification systems, coils/drain pans, or other HVAC components), some already prior suspects and some not, that when deficient may increase occupant symptoms, but that may be modified to prevent the symptoms. Other possible parts of the explanation include inadequate outdoor air ventilation (Seppanen et al., 1999) and inadequate thermal control (Mendell et al., 2002b). An additional possibility, not yet widely considered, is that, in general, conventional HVAC system design in air-conditioned buildings, involving frequently wet surfaces on cooling coils, drain pans, and sometimes humidifiers, may lead to as yet uncharacterized microbiologic exposures and consequent illness syndromes (Mendell, 2004; Menzies et al., 2003). Poor maintenance or condition of a specific HVAC or humidification system would then further increase a generally higher baseline risk associated with HVAC systems.

There is limited information in the associations found to suggest specific biologic mechanisms of response, based on increases in similar symptoms to related risks (e.g. potentially allergic symptoms associated with HVAC features related to moisture and microbiologic contamination). Low height of outdoor air intake was associated with almost all symptom outcomes assessed, including lower and upper respiratory symptoms, fatigue/difficulty concentrating, headache, and skin symptoms. Although vehicular pollutants seem to be a likely explanation, specific mechanisms to explain these symptoms are not obvious. Rubino et al. (1998), based on their study of vertical concentration gradients of outdoor vehicle-related pollutants, recommended that, as indoor air quality in urban buildings relies heavily on outdoor air quality, and large improvements in outdoor air quality are achievable only with large-scale urban changes, increasing height of outdoor air intakes may allow practical improvements in indoor air quality.

Humidification systems in poor condition were associated with increased upper respiratory symptoms, fatigue/concentration difficulty, eye and skin symptoms, and headache. Microbiologic contamination seems the most likely source of any resulting exposures, through allergic, irritant, or other mechanisms. Chemical biocides used in some humidifier systems, however, are also a potential explanation. On the other hand, infrequently cleaned cooling coils and condensate drip pans were associated with different symptoms – lower respiratory symptoms and headache – suggesting possibly different biologic mechanisms.

Overall, findings here and elsewhere suggest that some aspects of HVAC systems, related to contaminants either produced by the systems or brought in from outside, may increase risk of building-related symptoms in office buildings. The data used for these analyses, however, were not collected to explore our specific hypotheses, and thus were not optimal for that purpose. Future hypothesis-driven research, including detailed collection of appropriate data related to both environmental risk factors and health outcomes, is necessary to confirm and clarify the relationships reported here. If the relationships we have found are confirmed, then relatively straightforward modifications to the design and maintenance of ventilation systems may allow reductions in symptoms for millions of office workers.


This work was supported by the Indoor Environments Division, Office of Radiation and Indoor Air, Office of Air and Radiation of the US Environmental Protection Agency through interagency agreement DW8992169501-1 with the US Department of Energy. The work was also supported by the Finnish Academy. Conclusions in this paper are those of the authors and not necessarily those of the US Environmental Protection Agency. We thank Rick Diamond and Michael Bates for their review of the draft manuscript.


Appendix 1: Modeling procedures

All analyses were performed using sas version 8 (SAS Institute Inc, 2002). We first identified all variables in the BASE data corresponding to risk factors of interest for this analysis: features or practices in buildings hypothesized to be related to contamination or performance of HVAC system, including aspects of design, operation, maintenance, and condition of HVAC systems. We used data for the specific test space studied in each building where available and appropriate; otherwise we used data applicable to each entire building. Additional steps in the analyses were as follows:

  • • From univariate/descriptive analyses of potential risk variables, we excluded those with too many missing values (>10%), or insufficient variation in the key contrast (less than 5% of observations in any key category). We collapsed categories where necessary and feasible, and in some cases created combined variables or indices that summarized risks from closely related or highly correlated variables, to create the initial set of risk factor variables. Some of these changes were based on initiate bivariate models. All risk factor variables were dichotomous or categorical.
  • • For each of the seven symptom outcomes, we performed bivariate analyses with the initial risk factor variables, retaining for further analyses those with at least moderate associations. For a ‘moderate’ association, we required either an overall P-value < 0.25 or, for multicategorical variables, a P-value < 0.15 for any single category or for the Mantel–Haenzsel trend statistic. Starting with this step, the set of risk factors was allowed to differ between the different symptom outcome models.
  • • For each symptom outcome, we then examined all risk variables remaining after step 2 together in a multivariate logistic regression ‘risk selection’ model, to identify and omit variables with little association with the outcomes when adjusted for other risk factors. We sequentially excluded the variable with the highest P-value, stopping when all P-values were <0.20. We also identified highly correlated variables and combined, revised, or eliminated them as necessary.
  • • To the reduced set of risk variables left in each outcome model, we added potential confounding variables, personal or environmental. Potential confounders were added sequentially to the model and retained if an addition changed the point estimate for any risk factor by at least 10%.
  • • We then reconsidered previously rejected risk variables again for contribution to these expanded models, retaining any with a P-value < 0.05.
  • • Finally, we considered potential interactions between some variables in the models, but identified no evident important interactions.
  • • At this point, we examined the Hosmer–Lemeshow goodness-of-fit statistic, and if P < 0.05, we omitted the confounders with the highest P-values until the goodness-of-fit P-value ≥ 0.05.
  • • Using the set of final logistic regression models for each of the seven outcomes, we then imputed missing values on personal variables (but not environmental or building variables) using SAS Proc MI (four iterations) and reran the final logistic regression models. Finally, we used general estimating equations (GEE) in SAS Proc Genmod to adjust for potential correlation of observations within each building.



Abstract  Building-related symptoms in office workers worldwide are common, but of uncertain etiology. One cause may be contaminants related to characteristics of heating, ventilating, and air-conditioning (HVAC) systems. We analyzed data from 97 representative air-conditioned US office buildings in the Building Assessment and Survey Evaluation (BASE) study. Using logistic regression models with generalized estimating equations, we estimated odds ratios (OR) and 95% confidence intervals for associations between building-related symptom outcomes and HVAC characteristics. Outdoor air intakes less than 60 m above ground level were associated with significant increases in most symptoms: e.g. for upper respiratory symptoms, OR for intake heights 30 to 60 m, 0 to <30 m, and below ground level were 2.7, 2.0, and 2.1. Humidification systems with poor condition/maintenance were associated with significantly increased upper respiratory symptoms, eye symptoms, fatigue/difficulty concentrating, and skin symptoms, with OR = 1.5, 1.5, 1.7, and 1.6. Less frequent cleaning of cooling coils and drain pans was associated with significantly increased eye symptoms and headache, with OR = 1.7 and 1.6. Symptoms may be due to microbial exposures from poorly maintained ventilation systems and to greater levels of vehicular pollutants at air intakes nearer the ground level. Replication and explanation of these findings is needed.