Despite the shown enhanced use of MSI and intact selective attention, OA perform less well than YA in tasks that require more cognitive function than simple stimulus detection tasks. OA need more time to accurately perform more complex tasks in comparison to YA and show longer reaction times (e.g., selective attention task or space localization task; Furman et al., 2003; Diederich et al., 2008; Hugenschmidt et al., 2009b; Stephen et al., 2010; Elliott et al., 2011; Mahoney et al., 2011, 2012; Dobreva et al., 2012; Wu et al., 2012; DeLoss et al., 2013; Temprado et al., 2013; Guerreiro et al., 2014, 2015). In addition, OA are less accurate and more variable at performing tasks like navigation or localizing a target in space (Deshpande and Patla, 2007; Redfern et al., 2009; Berard et al., 2012; Bates and Wolbers, 2014; Deshpande and Zhang, 2014; Eikema et al., 2014; Hugenschmidt et al., 2014).

Additionally, OA were found to be impaired in properly weighing relevant and irrelevant sensory information from one’s own body and from the environment. Specifically, data suggest that, in comparison to YA, they do not properly adjust information that is unreliable (disrupted or taken away) or non-informative (distractors). They continue to use all environmental information when they should not (Stelmach et al., 1989; Teasdale et al., 1991; Strupp et al., 1999; Redfern et al., 2001, 2009; Allison et al., 2006; Poliakoff et al., 2006a; Deshpande and Patla, 2007; Hugenschmidt et al., 2009a; Elliott et al., 2011; Guerreiro and Van Gerven, 2011; Berard et al., 2012; Dobreva et al., 2012; Wu et al., 2012; DeLoss et al., 2013; Guerreiro et al., 2013; Bisson et al., 2014; Deshpande and Zhang, 2014; Eikema et al., 2014; McGovern et al., 2014; Brodoehl et al., 2015). These results were seen in several tests assessing integration of all combinations of modalities found in this review. These tests include the sound-induced flash illusion, n-back tasks with distractors, walking and navigation tasks, postural tasks with or without cognitive dual task, selective attention task with distractors, visual straight ahead tasks, control of movement timing tasks and visual-vestibular task with cognitive dual task. When the accuracy of a modality was restored in the trials, OA failed to use the correct information properly and as a result, they were even more perturbed while the YA adapted rapidly (Teasdale et al., 1991; Berard et al., 2012; Eikema et al., 2014). They are thus impaired in rapidly adapting their behavior to the environment which can be an issue for the performance of ADLs. As far as we know, the fact that OA are impaired at properly weighing sensory information has not been described earlier in the literature.

Dual tasking involves the concurrence of two different activities and requires high attentional demand. The results show that the addition of a dual task decreases the performance of both age groups, and that this effect is larger in the elderly population (Redfern et al., 2001, 2009; Mahboobin et al., 2007; Bisson et al., 2014). It seems that OA are unable to compensate for the increase in attentional demands and have difficulties to accurately perform multiple tasks at the same time.

A volume reduction of the temporal lobe has been hypothesized to be an anatomical cause of the changes in MSI found in the elderly population. Several brain areas have been found to contribute to the process of MSI, the impact of one sensory modality on the brain activity produced by another sensory modality. The STS and the superior colliculus (SC) have been found to be major actors of this process (Calvert and Thesen, 2004; Clemo et al., 2012). The STS and the SC have been shown to receive projections from areas involved in visual processes, auditory processes, and somatosensory processes (Clemo et al., 2012). Other regions of the brain have been found to be involved in multisensory processing, for instance, the claustrum, the suprageniculate and medial pulvinar nuclei of the thalamus and the amygdaloid complex (Calvert and Thesen, 2004).

After 35 years of age, brain volume starts to reduce (Hedman et al., 2012). While several parts of the brain are affected by this volume loss, the prefrontal cortex and the striatum are the most affected (Peters, 2006). The volumes of the temporal lobe, cerebellar vermis, cerebellar hemispheres and hippocampus are also decreased by age as well as the prefrontal white matter (Peters, 2006). The STS, which is highly involved in MSI processes as seen above, is situated in the temporal lobe of the brain.

The reduction of brain volume observed in the elderly population has been claimed to be the cause of major changes in OA’ capacities (Hedman et al., 2012) and changes on brain activation (Peters, 2006).

It has been shown that brains of OA tend to show more symmetrical activation than younger brains (Cabeza, 2002; Peters, 2006; Greenwood, 2007; Park and Reuter-Lorenz, 2009). This hemispheric asymmetry reduction in OA is called HAROLD (Cabeza, 2002). Different explanations have been explored to explain these findings. A failure to recruit the specific areas needed for the task and inhibition of the non-relevant areas, an attenuation of the response seen in YA or a compensation strategy of the aging process have been proposed (Peters, 2006; Park and Reuter-Lorenz, 2009). HAROLD was found to be correlated with higher performances in task execution in the elderly population, leading to the hypothesis that these changes occur to preserve the good functioning of cognition in OA (Greenwood, 2007; Park and Reuter-Lorenz, 2009). Additionally, during multisensory tasks, OA were shown to recruit more brain areas than YA (Townsend et al., 2006; Heuninckx et al., 2008; Venkatraman et al., 2010).

These changes of brain areas recruitment during MSI could serve as a compensation strategy for age-related deteriorations in individual sensory and motor systems and permit the elderly population to detect the stimuli as accurately as the YA (Cabeza, 2002). A dedifferentiation effect has also been proposed as an explanation (Baltes et al., 1980; Cabeza, 2002). Learning causes localized changes in specific areas of the brain needed for the task (Baltes et al., 1980; Bransford et al., 2000; Greenwood, 2007; Lövdén et al., 2013). Initially, several brain areas are recruited but as soon as the participant becomes an expert in the task, the expansion is followed by a renormalization of the activation map in which the most efficient circuits are selected (Lövdén et al., 2013). In the elderly population, this differentiation and specialization could be lost and OA start recruiting again a higher number of brain areas, the MSI control is unlearned (Baltes et al., 1980; Cabeza, 2002).

These anatomical modifications could be part of the changes that occur with aging regarding MSI by modifying the information processing in the brain of OA compared to YA. Although evidences of a link between anatomical changes and cognitive function have been described in the literature (Glisky, 2007), the exact nature of this relationship is not yet known and complex to investigate (Glisky, 2007).

MSI involves both top–down and bottom–up processes (Guerreiro et al., 2010; Talsma, 2015). MSI occurs pre-attentively in an automatic bottom-up process and is driven by the stimulus salience (Guerreiro et al., 2010; Talsma, 2015). The control of MSI is a top-down process driven by several components, expectations and goals for instance (Guerreiro et al., 2010; Talsma, 2015). However, an object integrated by more than one sensory system captures one’s attention more efficiently and proves that bottom-up integration can “drive” attention (Talsma, 2015). Additionally, the integration of stimuli depends on its relevance, for instance, a task-irrelevant sound associated with an attended visual stimuli will be more likely to be integrated compared to a task-irrelevant sound associated with an unattended visual stimulus (Guerreiro et al., 2010; van Erp et al., 2013; Talsma, 2015). Similar effects are found for visual and tactile stimuli (Philippi et al., 2008; Werkhoven et al., 2009; van Erp et al., 2014). These results show that top-down and bottom-up multisensory processes are closely interlinked. This systematic review shows that both top-down and bottom-up processes of MSI are affected by age. OA fail to use properly the bottom-up multisensory process of weighing information using their salience. Selective integration (top-down) is also hampered with age. As seen in the results above, OA are more affected than YA by dual tasks and especially cognitive tasks (Redfern et al., 2001, 2009; Mahboobin et al., 2007; Bisson et al., 2014).

Different theories have been explored to explain the results found in this systematic review in other reviews, particularly the increased use of MSI in the elderly population (Mozolic et al., 2012; Freiherr et al., 2013). These theories are explained below.

OA were usually slower and impaired in task performance, particularly when the task was cognitively demanding or more difficult (Furman et al., 2003; Diederich et al., 2008; Hugenschmidt et al., 2009b; Stephen et al., 2010; Elliott et al., 2011; Mahoney et al., 2011, 2012; Dobreva et al., 2012; Wu et al., 2012; DeLoss et al., 2013; Guerreiro et al., 2014, 2015). Mozolic et al. (2012) argued that a unisensory presentation of a stimulus is a more demanding task than the multisensory presentation of this stimulus because the multisensory task provides redundant information (same stimuli in different modalities).

It could then be assumed that the high multisensory gain shown in the elderly population would be caused by MSI being a less demanding task than using unisensory information (Mozolic et al., 2012). However, when general cognitive slowing was reduced by the use of a simple task such as an audiovisual detection task (Peiffer et al., 2007), the higher MSI gain was still visible in OA compared to YA. Thus, general cognitive slowing cannot explain by itself the differences in multisensory processing between OA and YA.

The inverse effectiveness is the principle that “decreasing the effectiveness of individual sensory stimuli increases the magnitude of multisensory enhancements” (Mozolic et al., 2012). It means that multisensory stimuli presented at a low level of salience (less intense or weak and ambiguous) are more likely to be integrated than unisensory stimulus presented at a high level of salience (Freiherr et al., 2013).

It is known that OA experience a functional decline in individual sensory systems (Teasdale et al., 1991; Bugnariu and Fung, 2007; Cruz-Jentoft et al., 2010; Owsley, 2011; Yeh et al., 2015). According to this principle of inverse effectiveness, this could lead to an increased multisensory benefit. However, in some studies included in this review, OA showed the same reaction times as the YA for unisensory stimuli (Townsend et al., 2006; Peiffer et al., 2007; Hugenschmidt et al., 2009a, b; Guerreiro et al., 2012, 2014, 2015; Fiacconi et al., 2013). This means that for some tasks, OA didn’t experience a functional decline effect in individual sensory systems compared to YA but still showed a multisensory facilitation. As a consequence, the inverse effectiveness cannot be the only process involved in the multisensory enhancement shown in the elderly population.

OA were found to have a “larger period for potential interaction” compared to YA as a consequence of broader distribution and increased response times (Peiffer et al., 2007; Diederich et al., 2008; Mozolic et al., 2012).

However, despite this larger time window of integration, increased reaction times and increased response variability actually reduce the probability of the overlapping of stimuli from different modalities in this time window (Diederich et al., 2008; Freiherr et al., 2013). Therefore, this hypothesis cannot explain why the use of MSI is higher in OA compared to YA (Mozolic et al., 2012; Freiherr et al., 2013).

Selective attention is the ability to focus on one stimulus or one modality while ignoring others (Mozolic et al., 2012; Freiherr et al., 2013). The brain activity of OA during selective attention for MSI has been shown to be different than the one of YA, who have an increased brain activity in areas associated with the attended modality and decreased brain activity in areas associated with unattended modalities (Mozolic et al., 2012).

Deficits in attentional control in the elderly population could then be assumed to take part in the increased amount of multisensory information being processed. OA fail to focus on one stimulus but rather integrate all the information available to them. However, several studies found that OA were still able to engage selective attention in simple tasks (Townsend et al., 2006; Hugenschmidt et al., 2009a,b; Guerreiro et al., 2012, 2014, 2015; Fiacconi et al., 2013). As a consequence, selective attention cannot solely explain the increased MSI in OA.

Nevertheless, deficits in selective attention could explain the fact that OA were more distracted by stimuli within the same modality or in another modality as the attended stimulus (Furman et al., 2003; Poliakoff et al., 2006a; Townsend et al., 2006; Diederich et al., 2008; Hugenschmidt et al., 2009a; Peiffer et al., 2009; Guerreiro and Van Gerven, 2011; Setti et al., 2011a; Guerreiro et al., 2012, 2013, 2014, 2015; DeLoss et al., 2013; McGovern et al., 2014).

None of the hypotheses described above are entirely able to explain the increased MSI in the elderly population (Mozolic et al., 2012; Freiherr et al., 2013). Mozolic et al. (2012) developed another hypothesis explaining the differences between OA and YA: increased noise at baseline. The authors argued that when OA engaged in selective attention, multisensory areas activity was reduced but remained higher than the YA, leading to sensory noise. When YA engaged in selective attention, the multisensory areas enhancements in their brain were suppressed to successfully ignore non-relevant information. The authors explained that because of this noise, OA were less able to ignore distractors but when information from the environment became relevant, they benefited from this higher baseline and showed larger MSI responses. This is beneficial when all information is reliable, and a disadvantage when part of the information should be ignored. This hypothesis could explain why OA maximized the use of MSI but are still impaired regarding their performance in the task or when the task is more difficult (e.g., cognitively demanding). This hypothesis fits best to the results described in the systematic review.

Five different hypotheses have been put forward by other authors to explain the differences between younger and OA regarding MSI: a general cognitive slowing, the inverse effectiveness, a larger time window of integration, deficits in attentional control and the one that fits best the results, the increased noise at baseline. However, this systematic review pointed to an age-related change that has not been described in the previous systematic reviews: a deficit in the weighing of sensory information in the elderly population compared to YA.

In this systematic review, we found that OA were impaired at properly weighing sensory information from the environment compared to YA (Stelmach et al., 1989; Teasdale et al., 1991; Strupp et al., 1999; Redfern et al., 2001, 2009; Allison et al., 2006; Poliakoff et al., 2006a; Deshpande and Patla, 2007; Hugenschmidt et al., 2009a; Elliott et al., 2011; Guerreiro and Van Gerven, 2011; Berard et al., 2012; Dobreva et al., 2012; Wu et al., 2012; DeLoss et al., 2013; Guerreiro et al., 2013; Bisson et al., 2014; Deshpande and Zhang, 2014; Eikema et al., 2014; McGovern et al., 2014; Brodoehl et al., 2015). The experiments that revealed this finding encompassed tests in which the sensory information was disrupted or taken away or when distractors were included. These age-related changes have not been described in the reviews that we found on age-related effects on MSI (Mozolic et al., 2012; Freiherr et al., 2013), probably because the previous reviews mostly focused on audiovisual tasks with a static position (Mozolic et al., 2012; Freiherr et al., 2013) while this effect was particularly observable when wrong information was presented during postural tasks involving visual, somatosensory and vestibular information. For example, Bates and Wolbers (2014) showed that OA relied less than predicted on visual landmarks in a navigation task. This was also seen when a GVS was added to help the subjects reduce postural sway: OA were unable to properly use the added information (Eikema et al., 2014).

The changes in the weighing of the information could be caused by a failure in detecting that the information is important or unreliable, and/or in a failure in inhibiting the use of unreliable information. This would be consistent with Mozolic et al. (2012) hypothesis, the increased noise at baseline in the elderly population leads to sensory noise and could hinder them judging if the information is irrelevant or unreliable. OA were shown to recruit more multisensory brain areas than YA (Townsend et al., 2006; Heuninckx et al., 2008; Venkatraman et al., 2010), specifically frontal areas (Freiherr et al., 2013). These areas are known to be related to the selection of multisensory stimuli (Talsma, 2015) and connected to each other, leading to difficulties to downregulate individual modalities and irrelevant information (Berard et al., 2012; Mozolic et al., 2012). Besides, using all information available could be a good strategy for OA in whom one or more sensory sources have become unreliable due to bad unimodal processing, as this strategy could compensate for lower level sensory degradation (Berard et al., 2012).

The brain process of weighing sensory information from the environment follows the principle of Bayesian integration (Ernst, 2006; Bates and Wolbers, 2014; Ursino et al., 2014). This process aims to increase the accuracy of the percept by reducing its uncertainty (Bates and Wolbers, 2014). Stimulus information comes to a person through different modalities, for instance, the size of an object can be estimated through vision and haptics. The Bayesian model assumes that the brain weighs each signal optimally with respect to its variance and combines them into one estimate with a smaller variance than the variance of the individual estimates (Ernst, 2006; Ursino et al., 2014). According to the maximum likelihood estimation, the reliability of the combined estimate is the sum of the individual estimates (Ernst, 2006), i.e., it is generally valuable to integrate stimuli from different modalities as OA seem to do.

It has been shown that the human brain seems to integrate cue information in a Bayesian optimal manner. Computational studies have been done considering different areas of the brain (Anastasio et al., 2000; Colonius and Diederich, 2004b; Gu et al., 2008; Ursino et al., 2014). The authors found that the neurons present in the SC and the dorsal medial superior temporal area use the Bayesian rule to integrate and weigh multisensory information to arrive at a least variable percept.

The results described earlier suggest that OA are impaired at properly weighing sensory information from the environment. Additionally, Bates and Wolbers (2014) showed that OA relied less than optimally expected on visual information in a navigation task by using Bayesian modeling as described above. Age-related changes in the brain, for instance, gray or white matter losses (Peters, 2006; Hedman et al., 2012) or differences in brain activity (Cabeza, 2002; Peters, 2006; Greenwood, 2007; Park and Reuter-Lorenz, 2009) could have led to this degradation in the integration of the sensory cues present in the environment of OA. However, this level of integration could still be optimal for the performance of elderly people given their degradation of unisensory perception (Teasdale et al., 1991; Bugnariu and Fung, 2007; Owsley, 2011).

These findings will need further investigation with new experiments in order to better understand age-related changes in MSI in the healthy elderly population.