|Home | About | Journals | Submit | Contact Us | Français|
Visual masking procedures assess the earliest stages of visual processing. Patients with schizophrenia reliably show deficits on visual masking, and these procedures have been used to explore vulnerability to schizophrenia, probe underlying neural circuits, and help to explain functional outcome.
The goal of the current study was to identify and compare regional brain activity associated with one form of visual masking (i.e. backward masking) in schizophrenia patients and healthy controls.
Nineteen patients with schizophrenia and 19 healthy control subjects were recruited from the VA Greater Los Angles Healthcare System and local communities. Subjects received functional magnetic resonance imaging (fMRI) scans. While in the scanner, subjects performed a backward masking task, and were given three functional localizer activation scans to identify early visual processing regions of interest (ROIs).
The main outcome measure was the magnitude of the fMRI signal during backward masking.
Two ROIs (lateral occipital complex (LO), and motion sensitive area, hMT+) showed sensitivity to the effects of masking, meaning that signal in these areas increased as the target became more visible. Patients had lower activation than controls in LO across all levels of visibility, but did not differ in other visual processing ROIs. Using whole brain analyses, we also identified areas outside the ROIs that were sensitive to masking effects (including bilateral inferior parietal lobe and thalamus), but groups did not differ in signal magnitude in these areas.
The study results support a key role in LO for visual masking, consistent with previous studies in healthy controls. The current results indicate that patients fail to activate LO to the same extent as controls during visual processing regardless of stimulus visibility, suggesting a neural basis for the visual masking deficit, and possibly other visual integration deficits, in schizophrenia.
Much can be learned about schizophrenia as early as the first tenth of a second of visual processing. In a visual masking paradigm, the visibility of a visual target is disrupted by the presence of a mask that occurs briefly before or after the target 1, 2. Many studies over the years have demonstrated that schizophrenia patients perform more poorly than comparison subjects in identifying the target in the presence of a mask 3–9. Visual masking in schizophrenia may provide a window into neural substrates, clinical symptoms, and vulnerability. Also, the degree of impairment may be associated with community functioning 10.
The deficits in backward masking (i.e. masking in which the mask appears after the target) in schizophrenia have generated interest for several reasons. First, performance on visual masking procedures may be a trait marker that reflects vulnerability to schizophrenia. Masking deficits have been reported in patients who are in clinical remission 11, 12 and, to a reduced degree, in first-degree relatives of patients 13–17, suggesting that visual masking deficits are related to genetic vulnerability to illness. Second, they represent a narrowly defined early perceptual process for which effortful attentional modulation has minimal effects 18. We have found visual masking impairment even after individuals are equated for sensory input of the target, and the groups did not differ in the contrast levels needed for target detection 19. Third, backward masking deficits have been related to clinical aspects of schizophrenia, including negative symptoms 20, 21. Fourth, initial evidence suggests that deficits on backward masking tasks are related to functional outcome, probably through key mediating variables. For example, visual masking has been found to be associated with performance on social cognitive tasks, which in turn are related to community functioning 10. Similar findings have been reported for other early visual processing tasks in schizophrenia 22.
A fifth reason for interest in backward masking paradigms is their detailed linkages to underlying neural systems 1, 2. The visual masking deficits have been viewed in terms of the interactions of two key visual channels that form the basis for complex visual processing and are labeled in functional (transient and sustained) or anatomical (magnocellular and parvocellular) terms 23. The magnocellular (“magno”) pathway is characterized by fast-phasic (transient) responses, while the parvocellular (“parvo”) pathway has slower-tonic (sustained) responses. Thus, magno channels respond rapidly and briefly to stimulus onset, offset, and location, whereas parvo channels have a longer response duration related to stimulus identification. According to this model of masking, parvo activity elicited by the target conveys information needed to identify targets and masking results from disruption of this target information by either the magnocellular or parvocellular pathways of the mask, depending on the particular paradigm 1, 24. Deficits in masking in schizophrenia have been attributed to abnormalities in the magnocellular pathway 3, 4, 25, 26. However, there is also interest in possible schizophrenia-related abnormalities in high frequency brain electrical activity (i.e., gamma range, above 30 Hz) that is associated with formation of visual representations 27, 28. In backward masking models, this gamma activity is presumed to occur in the parvocellular system 29.
Although the very brief temporal demands of backward masking present challenges for its application in the scanner, several recent studies have successfully examined masking paradigms with functional Magnetic Resonance Imaging (fMRI) in healthy samples (e.g. 30–33). While fMRI does not have the temporal resolution to examine the blood-oxygen-level dependent (BOLD) response to the target and mask separately, it can yield an informative BOLD response to the target-mask sequence. There have been no studies, to our knowledge, on the neural basis of visual masking in schizophrenia using any functional neuroimaging procedure. Hence, the regional brain activity associated with these key deficits is not known. We previously examined visual masking in a group of healthy subjects to determine which areas are sensitive to masking effects (i.e., show more activation with increasing duration between target and mask) 34. Among the early visual processing areas identified a priori through localizer scans, the lateral occipital complex (LO) showed strong sensitivity to masking effects. Hence, LO was of particular interest as a possible substrate of the masking effect. Other areas showed similar effects on post hoc analyses, including inferior parietal, anterior cingulate, and thalamus.
For the current study we used an approach similar to our previous study 34 in which we first examine well-defined early visual processing areas for which there are established localizer tasks. We refer to these regions as “a priori” regions of interest (ROIs). Because areas outside of these early visual regions may be involved with masking, and may explain group differences, we also conducted an exploratory whole brain analyses (i.e. post hoc analyses). We included samples of patients with schizophrenia (n = 19) and healthy controls (n = 19) to address several questions: 1) among the visual processing areas identified a priori as ROIs, which show sensitivity to the effects of visual masking? 2) of these ROIs, which show differences between patients and controls in level of activation? 3) do other areas (identified post hoc) show sensitivity to masking effects? and 4) do patients and controls differ in activation on any of these areas?
Twenty-four (5 female) patients with schizophrenia and 19 (5 female) healthy controls were recruited for this study. Participants were recruited from a larger NIMH study of early visual processing in schizophrenia 4, 5, 15. Schizophrenia patients were recruited from outpatient treatment clinics at the Veterans Affairs (VA) Greater Los Angeles Healthcare System and from local board and care facilities. Schizophrenia patients met diagnostic criteria for schizophrenia using the Structured Clinical Interview for DSM-IV Axis I Disorders 35. Exclusion criteria for patients included: 1) substance abuse or dependence in the last six months, 2) mental retardation based on review of medical records, 3) history of loss of consciousness for more than one hour, 4) an identifiable neurological disorder, or 5) not sufficiently fluent in English. We did not exclude patients for a history of substance dependence.
Normal control participants were recruited through flyers posted in the local community, newspaper advertisements in local newspapers, and website postings. Exclusion criteria for control participants included: 1) history of schizophrenia or other psychotic disorder, bipolar disorder, recurrent depression, history of substance dependence, or any substance abuse in the last 6 months based on the SCID 35, 2) any of the following Axis II disorders: avoidant, paranoid, schizoid, schizotypal, or borderline, based on the SCID for Axis II disorders 36, 3) schizophrenia or other psychotic disorder in a first-degree relative, 4) any significant neurological disorder or head injury, and 5) not sufficiently fluent in English.
Schizophrenia patients and normal controls were comparable in terms of age and education (age, 38.2 ±11.8 and 42.7 ±9.0, respectively, t36=−1.33, NS; education, 13.6 ±1.6 and 13.2 ±1.3, respectively, t36=0.95, NS). All patients were clinically stable at time of testing and exhibited low levels of symptoms (for a more complete characterization of the patient sample for the larger project, see 10). All of the patients were medicated (1 - conventional antipsychotic, 16 - single atypical, and 2 - two atypical antipsychotic medications). Subjects did not take any sedatives during the 12 hours prior to assessment. All participants had normal or corrected to normal vision, and we did not select subjects on handedness.
All SCID interviewers were trained to a minimum kappa of 0.75 for key psychotic and mood items through the Treatment Unit of the Department of Veterans Affairs VISN 22 Mental Illness Research, Education, and Clinical Center (MIRECC). None of the research staff was blind to group membership. All participants were evaluated for the capacity to give informed consent and provided written informed consent after all procedures were fully explained, according to procedures approved by the Institutional Review Boards at UCLA and the VA Greater Los Angeles Health Care System.
All participants completed six runs of the visual backward masking task followed by three localizer tasks (LO, retinotopic areas, and hMT+) in the scanner. The entire scanning session lasted 60 minutes. The visual backward masking task was presented using E-prime software and the localizer tasks were presented using MATLAB. All tasks were presented with MR-compatible LCD goggles (Resonance Technology, Northridge, CA).
For the visual backward masking task, the target was a square with a gap on one of three sides (up, down, or left) that appeared at the center of the screen. The mask was a composite square made up of four smaller squares, overlapping the area occupied by the target (see Figure 1). A mask was presented at one of four possible stimulus onset asynchronies (SOAs). These SOAs and stimuli durations were decided upon after piloting with the MR-compatible goggles. The beginning of each visual backward masking trial was signaled by two 100 ms flashes of a fixation point, followed by a 600-ms blank period. The target duration was held constant at 26.6 ms, and followed by a 53.3-ms mask at one of four possible SOAs: 26.6, 40, 80, 200 ms. The target subtended 5.7 degrees and the mask 10.2 degrees of visual angle. Participants were instructed to identify the location of a gap in the target (up, bottom, or left) by pressing a corresponding button with their dominant hand.
The visual backward masking tasks consisted of 6 runs, each with 30 5-second trials (i.e. 6 trials for each of the 4 SOAs and 6 null trials that included fixation but no stimuli). We employed a rapid event-related design and the trials were presented in a “permuted block design” to maximize both hemodynamic response function (HRF) estimation and signal detection power 37–39. To create trial sequences for each run, we first determined a random ordering of blocks, in which blocks contained three trials from a single condition (i.e., one of the four SOAs or null). Then, we randomly swapped trials in the sequence until the first order entropy of the sequence increased by a factor of 1.33. This procedure made the trial sequence appear random, and it was not easy for subjects to form expectations about the upcoming trial type. To generate the final sequence, this process was repeated 500 times and the trial sequence with the highest statistical efficiencies were selected 40.
After the visual backward masking task, participants performed three functional localization tasks: LO, retinotopic areas, and hMT+. Full descriptions of the three functional localizer tasks are provided elsewhere 41 and are summarized briefly here. The LO localizer task consisted of alternating blocked presentations of pictures of abstract objects (i.e. sculptures) and scrambled pictures of objects, with each block containing 10 images presented for a total of 12.5 s 42, 43. The scrambled pictures were created by dividing each picture into 64 squares and randomly scrambling their positions. There were 6 blocks each of abstract objects and scrambled objects. Our definition of LO follows that of Malach and colleagues 44, and includes lateral cortex between V3 and MT+ but also continues into adjacent ventral cortex. To identify retinotopic areas, participants viewed slowly rotating wedges of a contrast-reversing checkerboard 45. The wedge made 5 rotations, with one rotation every 30 s. Retinotopic areas that respond well to the standard fMRI techniques for measuring retinotopic organization include V1, V2, ventral V3 (or VP), V3a and V4. The motion sensitive hMT+ localizer task consisted of alternating blocked presentations of moving rings and stationary rings, with each block presented for 15 s. There were 5 blocks each of moving and stationary rings.
All scanning was conducted on a 3T scanner (Siemens Allegra, Erlangen, Germany) located in the UCLA Ahmanson Lovelace Brain Mapping Center. For anatomical reference, a high-resolution Echo Planar axial T2-weighted series was obtained for each subject prior to functional scanning (TR = 6000 ms, TE = 54 ms, flip angle = 90 degrees, 30 axial slices, FOV 20 cm). A T2*-weighted gradient-echo sequence was used to detect blood-oxygen level-dependent (BOLD) signal (TR=2000 ms, TE=42 ms, flip angle=80 degrees, voxel size of 3.125 × 3.125 × 4.00 mm with a 1-mm gap), acquiring 24 slices parallel to the AC-PC plane.
A detailed description of the fMRI data analysis is included in supplemental material. In brief, we approached the fMRI data analyses in two complementary ways. First, we used an ROI-based approach in which we: 1) identified three key visual processing areas separately for each subject using the functional localizer tasks, 2) examined the effect of SOA (that is, masking effect) for each ROI, and 3) examined between-group differences at each ROI. Second, we conducted a whole brain analysis in which we: 1) identified clusters of voxels that were sensitive to SOA (masking) effects outside the 3 ROIs, and 2) compared the two groups on these areas. Analyses were performed using automated procedures that do not involve user-adjusted parameters; thus the experimenter was not blinded to the subjects’ group assignments.
Five patients were excluded from analyses: one subject had excessive movement artifact due to tardive dyskinesia, two were excluded due to technical problems during scanning, and two showed chance-level performance (defined as at or below 38% accuracy) at the longest SOA (200ms). Therefore, 19 schizophrenia patients (4 females) and 19 normal controls (5 females) were included in the following analyses.
Behavioral data across SOAs are shown in Figure 2. A repeated measures ANOVA yielded a significant effect of SOA (F1,36 = 350.9, p<.0001). Neither the main effect of group nor the group by SOA interaction was significant. Performance was comparable at all points except SOA 3. Controls performed .50 effect size better than patients at this SOA but the difference was not significant at these sample sizes. As expected, both schizophrenia patients and normal controls showed improved performance as SOAs increased (i.e., the masking became weaker). Because the performance data for SOA 1 and SOA 2 were close to chance performance for both groups, we combined the responses for these SOAs in all subsequent analyses.
There were no significant differences (all p values > .20) between groups in the number of voxels for any of the ROIs (See Figure 3: Retinotopic areas: controls mean=360, SE=46, patients mean=380, SE=50; hMT+: controls mean=57, SE=11, patients, mean=59, SE=12; LO: controls mean=34, SE=12, schizophrenia patients, mean=56, SE=11).
The time series of signal change for each ROI are presented for patients and controls in Figures 4–6. Our initial question was whether any of the ROIs showed sensitivity to masking (i.e., an SOA effect at the middle time points) across groups. We conducted a 3 (SOA) by 7 (time points) by 2 (group) ANOVA separately for each ROI. Our interest was whether there was: 1) an SOA by time point interaction or 2) an SOA by group interaction. Each ROI had a significant SOA by time point interaction (F 12,396=2.54, p<.05, F 12, 384=1.92, p<.05, F 12,384=2.68, p<.05 for retinotopic areas, hMT+, and LO, respectively). None of the ROIs had a significant SOA by group interaction, so we combined groups and conducted post-hoc ANOVAs at each time for each ROI to better understand the SOA by time point interaction. For retinotopic areas (Figure 4), there was no significant effect of SOA at any time point. For hMT+ (Figure 5), significant SOA effects were present at time points 6 and 8 (F 2,66=3.23, p<0.05, and F2,66=4.04, p<0.05, respectively). For LO (Figure 6), we also found significant SOA effects at time points 6 and 8 (F 2,66=6.49, p<0.01, and F 2,66=5.28, p<0.01, respectively).
While the SOA effect for LO was predicted, the one for hMT+ was not. To determine whether the SOA effect for LO was greater than that of hMT+, we conducted a 3 (SOAs) × 2 (LO and hMT+) repeated measures ANOVA, separately for time points 6 and 8. For time point 6, we found significant main effects of ROI (F1,29=10.40, p<0.01) and SOA (F2,58=5.49, p<0.01). Notably, the ROI by SOA interaction was significant (F2,58=3.27, p<0.05). Hence, LO showed more activation overall, and a clearer SOA effect (more activation with increasing visibility of the target) compared with hMT+ (see figs 5 and 6). For time point 8, we found a significant effect of ROI (larger for LO, F1,29=7.77, p<0.01) and a significant effect of SOA (F2,58=4.81, p<0.05), but not a significant ROI by SOA interaction.
Next we considered any differences between groups by comparing their peak amplitude of percent signal change (i.e., peak signal change) for each ROI (shown in Figure 7). For retinotopic areas, and hMT+, none of the main effects or interactions was significant. The groups were generally comparable in peak signal change for these ROIs (p values = .59, and .37 for retinotopic and hMT+, respectively). For LO, we found a significant group effect (F 1,32=11.74, p<0.01), a significant effect of SOA (F2,64=3.42, p<0.05), and the SOA by group interaction was not significant. Hence, patients showed lower peak signal change on LO across SOAs, but a similar SOA effect, compared with controls.
Some authors have separated out the dorsal and ventral portions of LO, with a particular interest in ventral LO for object representation 34, 46. As an additional check on the findings, we examined the ventral portion of LO only. The results were essentially the same as for the entire LO: this region showed a clear effect of SOA and the patients showed lower amplitude responses. Figures for ventral LO are available as Supplemental Figure S1.
For the whole brain analyses, we first identified voxels that showed a parametric increase of SOA 1,2 < SOA 3 < SOA 4. With both patients and controls combined, the activated brain regions include: posterior cingulate cortex, right postcentral gyrus, bilateral inferior parietal lobe, bilateral superior temporal gyrus, bilateral middle temporal gyrus, left inferior temporal gyrus, bilateral fusiform gyrus, and thalamus (see Table 1). An image of the relevant brain images is available as Supplementary Figure S2. A second analysis compared activity levels in patients and controls, but found no reliable regions of differential activation.
The current study used two different approaches (an a priori ROI approach with localizer scans, and whole brain analyses) to examine the neural basis of visual masking in schizophrenia. Of the visual processing areas that were defined a priori through localizer scans, both LO and hMT+ showed sensitivity to the effects of masking, meaning that the activation increased as the SOAs become longer and the target more visible. The masking effect was significantly more pronounced in LO than in hMT+ at the 6 second time point. Compared with controls, patients showed lower peak signal change in LO, but not in the other visual processing areas. We identified areas outside of the visual processing areas that were sensitive to the effects of masking, including some areas that were implicated in our previous study (bilateral inferior parietal lobe and thalamus), but the groups did not differ in their activation of these areas.
The results of this study provide additional support for a key role in LO for visual masking, which is consistent with our previous study 34 and with other fMRI studies of visual masking in controls 31, 32. The current results also indicate that patients fail to activate LO to the same extent as controls across SOAs. The groups showed a non-significant .50 effect size difference on SOA 3, so we computed the correlation between accuracy and peak signal change on LO at this SOA in each group. The correlation was moderate, but non-significant in normal controls (r= .34), and small in schizophrenia patients (r= .09).
The blunted LO response was a general effect, observed for both short SOAs, where masking rendered the stimulus almost invisible, and the longest SOA, where masking had little effect. This pattern suggests that patients have generally weaker stimulus representations. Such a response from LO could not be inferred from behavioral masking studies of schizophrenia, in which it is common for patients to differ from controls at mid-range SOAs (ie. those without floor and ceiling effects). However, with fMRI it is possible to see a deficit across a range of visibility. Because of its generality, the blunted LO response could reveal itself as reduced performance in other tasks that depend heavily upon precise read-out of LO activity. Schizophrenia is associated with impairment on a range of visual integration tasks, including object recognition, grouping, perceptual closure, face processing, and reading 47, and deficits on any of these could be associated with a blunted LO response.
Other studies have used electrophysiology to suggest LO abnormalities in schizophrenia 48, 49. For example some have examined an event-related potential, closure negativity (Ncl), that is associated with object recognition. This wave has peak amplitude at approximately 290 ms and is distributed over the lateral occipital scalp regions. Hence, its distribution would include region LO assessed in the current study. Reports have found that this wave differs in magnitude 48 or pattern 49 in patients with schizophrenia, thereby providing indirect, but consistent, support for LO abnormalities in schizophrenia.
What processes might be occurring in LO? One probable function of LO is to integrate different types of information from a variety of sources to provide a unified representation of visual shape 50. The reduced LO activity we observed in patients may indicate that patients have a weaker unified representation, at least under conditions of masking. Because LO dysfunction appears to be a key neural source of masking effects (e.g. 30–33), we propose that patients’ weakened LO representations underlie the masking impairment observed in laboratory studies. It is possible that other brain areas supply a compensatory response for the weakened LO representation, however, the whole brain analyses did not reveal group differences in regions outside LO that would suggest such compensation.
Patients increased susceptibility in masking could originate from either feed-forward or reentrant processing in LO. Visual masking has typically been interpreted in terms of feed-forward models of visual processing (i.e., involving the flow of information from retina to cortex). In contrast to this model, more recent advances in cognitive neuroscience argue for the importance of reentrant (top-down or attentional) processes in early components of visual processing 51, 52. Reentrant models of visual processing emphasize the iterative processing of visual stimuli within the cortex. One recent study used fMRI to examine a type of backward masking (4-dot masking) that is thought to involve primarily masking by reentrant processes 32. The results supported the idea that reentrant processes disrupted the neural representation of the target in LO. Hence, LO appears to be a critical site for neural representation of the target. Although our masking methods were not designed to separate feed-forward from reentrant processes, it is entirely possible that group differences in LO could be attributable to disruption in reentrant processes.
In the present study, we intentionally defined each ROI in a relatively restricted way, based on the largest single cluster for each activation task for each individual. This is a conservative approach that yields smaller, but hopefully more functionally homogeneous, regions. A previous publication from our laboratory 53 used a more liberal approach to define ROIs to examine the extent of activation with the localizer tasks in the absence of any visual masking procedures. In that study, patients differed from controls in having a significantly broader area of activation with the LO localizer stimuli, but not with retinotopic or hMT+ localizers, again suggesting that LO is a critical area to consider regarding group differences in visual perception. To ensure that the present findings were not a result of using a more restricted approach for defining ROIs, we reanalyzed the fMRI data using ROIs defined as in our previous study and found the same pattern of results with visual masking (i.e. reduced LO activity in schizophrenia patients).
Patients and controls did not differ in performance in the scanner. On the surface, this appears to be inconsistent with performance data from many previous studies. It should be emphasized that the methods we have used in our published performance studies were optimized to show group differences. The target stimuli are small (.24 degrees of visual angle) and are shown at visual threshold so that all of the subjects are equated for initial sensory input. The backward masking methods in the present report were optimized to show a BOLD response. The stimuli were large (target = 5.7 degrees, mask = 10.2 degrees), dark with high contrast, and well above threshold for identification. Hence, the target in this fMRI study was roughly 24 times as large as in our performance lab studies. In addition, we used fewer trials and assessed fewer SOAs than in our typical behavioral paradigms. Although the absence of large performance differences is unusual in the context of the masking literature, it provides an interpretive advantage in that performance level is not a confound in considering group differences in regional brain activity.
It remains possible that the lack of significant performance difference indicates that the patients in this study are atypical. To examine this possibility, we evaluated data from a subset of patients and controls who were also examined on visual masking performance out of the scanner as part of the larger masking protocol. Although the small number of subjects (10 patients, 17 controls) precludes formal statistical analysis, The between-group effect sizes for a similar masking task (i.e. high energy identification) of these subjects were 0.05, 0.31, 0.61, 0.73 for SOAs = 100, 125, 150, 225 ms, respectively. These effect sizes between groups were highly consistent with what is found in other studies when patients and controls are matched for sensory input values.. Hence, we do not consider the patient sample to be unusual in this regard.
Although fMRI does not lend itself well to separate examination of the parvo- and magnocellular visual pathways, limited inferences can be made based on the input of these pathways to visual processing regions 23. MT is considered part of the dorsal pathway which receives predominant input from the magno system, whereas LO is considered part of the ventral pathway and receives input from both magno and parvo systems 54. Hence, the current findings of abnormalities in LO are not inconsistent with previous evidence for a magno abnormality in schizophrenia 3, 4, 25, 26.
Assuming that LO is a critical site for neural representation of a visual target, and that it underlies visual masking, as well as other visual processing, deficits in schizophrenia, our data raise questions about other consequences of a dysfunctional LO. For example, could abnormalities in this region contribute in part to the functional disability in schizophrenia? This view, while clearly speculative, would fit with behavioral performance findings that early visual processing tasks are associated with performance on social cognitive tasks, which are related to community functioning in schizophrenia 10, 22. Hence, with sufficient samples it will be possible to test these speculations by using brain activation in selected regions of interest as key variables in proposed pathways that run from brain to perception to basic and social cognition and eventually to functioning in schizophrenia.
These figures show the time series of hemodynamic response (top two panels) and the mean peak signal change (bottom panel) for the ventral portion of LO for patients and controls.
This figure shows the results based on the combined sample of patients and controls from an exploratory whole brain analysis of regions that showed a parametric increase with increased SOAs (SOA 1,2 < SOA 3 < SOA 4). The coordinates of the regions are presented in Table 1.
Funding/Support: Support for this study came from NIMH Grants MH43292 and MH065707 (PI: Michael F. Green, Ph.D). The authors wish to thank Poorang Nori and Alisa Malin for assistance in data collection.
For generous support of the UCLA Brain Mapping Center, we also thank the Brain Mapping Medical Research Organization, Brain Mapping Support Foundation, Pierson-Lovelace Foundation, The Ahmanson Foundation, William M. and Linda R. Dietel Philanthropic Fund at the Northern Piedmont Community Foundation, Tamkin Foundation, Jennifer Jones-Simon Foundation, Capital Group Companies Charitable Foundation, Robson Family, and Northstar Fund.
Author contributions: Dr. Green takes responsibility for the integrity of the data and the accuracy of the data analyses. All authors had full access to all of the data in the study.
Financial disclosure: None of the authors have any to disclose.