Sleeping mattress determinants and evaluation: a biomechanical review and critique

Background Sleeping mattress parameters significantly influence sleeping comfort and health, as reflected by the extensive investigations of sleeping support biomechanics to prevent sleep-related musculoskeletal problems. Methodology Herein, we review the current trends, research methodologies, and determinants of mattress biomechanics research, summarizing evidence published since 2008. In particular, we scrutinize 18 articles dealing with the development of new designs, recommendation criteria, instruments/methods of spine alignment evaluation, and comparative evaluation of different designs. Results The review demonstrated that mattress designs have strived for customization, regional features, and real-time active control to adapt to the biomechanical features of different body builds and postures. However, the suggested threshold or target values for desirable spine alignment and body pressure distribution during sleep cannot yet be justified in view of the lack of sufficient evidence. Conclusions It is necessary to formulate standard objectives and protocols for carrying out mattress evaluation.


INTRODUCTION
Some of the most frequent postures and locomotions in our lifetime are those adopted during sleeping. Therefore, sleeping systems and supports are considered to be important environmental components affecting physical comfort during sleep and thus influencing health. However, sleep disorders and problems are common, resulting in poor work efficiency, absenteeism, and accidents (Swanson et al., 2011). Additionally, poor sleep quality can cause musculoskeletal problems, including chronic pain, low back pain, and arm-shoulder pain (Cimmino, Ferrone & Cutolo, 2011;Mork et al., 2013).
Although the relationship between biomechanics and sleep-associated musculoskeletal problems has been recognized, most mattress evaluations pertain to the use of insecure techniques such as subjective feedback or questionnaires (Tonetti, Martoni & Natale, 2011;Verhaert et al., 2011b). The subjective responses gathered from questionnaires may change after an adaptation period (Liu, Lee & Liang, 2011), and the outcome of subjective comfort evaluation is argued to be easily manipulable based on manufacturer demands because of the highly diverse and vague definition of "comfort" (Wu, Yuan & Li, 2018). To introduce more order into this field, the European standardization body published the EN 1957 standard as a guide to mattress firmness selection according to hardness and area under the deflection curve (European Committee for Standardization, 2000). However, the corresponding recommendations are based on subjective perception and supine posture only (López-Torres et al., 2008) and lack supporting experimental data.
Besides subjective feedback, sleep therapists identified spine and pressure point positions as additional decisive factors for prescribing sleeping support (Esquirol Caussa et al., 2017). In general, an ideal sleeping support system should maintain the spine physiologically aligned by providing adequate and appropriate mechanical support (DeVocht et al., 2006;Verhaert et al., 2011b). In contrast, the use of sagging supports is believed to result in discomfort, poor sleeping quality, and waking symptoms caused by the irritation of adjacent soft tissue and nerves of the distorted spine (Verhaert et al., 2011b). Moreover, the body-mattress interfacial contact-pressure relationship, which is recognized as an indicator of tactile comfort, should also be taken into account (Pearson, 2009), for example, the use of rigid surfaces results in high and concentrated pressure, which induces discomfort and poor localized blood supply and can therefore place bed-borne patients at risk of pressure ulcer formation (Berlowitz & Brienza, 2007).
Although sleeping mattress biomechanics have not been extensively studied, the corresponding works employed broadly variable investigation methods and determinants (Radwan et al., 2015). Herein, we aim to review and summarize the biomechanical methods and determinants used to evaluate domestic mattress design and selection criteria and thus hope to stimulate the formalization of objective standards for the biomechanical evaluation of sleeping supports.

SURVEY METHODOLOGY Search strategy
Article search was conducted using "sleeping," "mattress," and "biomechan Ã " as keywords. Several databases were accessed, including Medline, Pubmed, ScienceDirect, Cochrane Collaboration, Scopus, and the Web of Science. A gray literature search was conducted using Google Scholar. Studies conducted in 2008-2018 and published in peer-reviewed journals in English were considered.

Study selection
Article relevancy was determined through primary title-based screening by two independent reviewers, and a second abstract-and full text-based screening was conducted by the same reviewers. In case of discrepancy, two more reviewers were intervened, and the selection outcome was settled by discussion. Articles were excluded if they did not account for any biomechanical measurements or investigations, for example, some excluded articles only dealt with posture behavior, cardiovascular and pulmonary measurements, questionnaires, temperature, and humidity. Articles on medical-use mattresses, particularly those describing anti-pressure sore mattresses, were also excluded.

Data extraction
Details of the 18 articles qualified for this review are summarized in Table 1 (selection of population and mattress samples), Table 2 (methods and biomechanical outcome measures), and Table 3 (study design and research scope).

Subjects and population
In the reviewed articles, all participants were normal subjects, since the review was confined to the use of domestic mattresses. Subjects with sleep disorders, related medications, or musculoskeletal disorders, for example, spine and back pain, were excluded. However, no study addressed the assessment methods of these criteria. While we assumed that the exclusion process was based on self-reports, both sleep disorders and musculoskeletal problems could be often undiagnosed (Behar et al., 2013;Woolf, Erwin & March, 2012).
Two studies investigated only male subjects, and one study investigated only female subjects. No study attempted to compare the differences between males and females, despite the fact that gender can influence body built, joint stiffness, and comfort perception.
Although young adults were commonly recruited, López-Torres et al. (2008) compared the outcomes obtained for young adults and elderly people, revealing the absence of significant differences between these groups in terms of subjective perception. The works of Esquirol Caussa et al. (2017) and Palmero et al. (2017) encompassed a wide age range (4-93) of subjects, but the influence of age was not discussed in either case.
Considering the influence of body built, Esquirol Caussa et al. (2017) identified five body morphotypes based on body girth to determine individualized sleeping system recommendations. Additionally, body dimensions were used to estimate the customized configuration of mattress zonal stiffness, although its influence on the biomechanics outcome was not documented (Verhaert et al., 2011b(Verhaert et al., , 2012a. Finally, several studies confirmed that the required mattress stiffness depended on body form and spinal curvature (Yoshida, Kamijo & Shimizu, 2012;Zhong et al., 2014).

Mattress samples
Among the 18 articles, eight focused on market-available ordinary mattresses (Table 1), for example, Esquirol Caussa et al. (2017) explored mattress-topper-pillow combinations with different material densities (although the use of pressure units (kPa) for density quantification was not explained), while Yoshida, Kamijo & Shimizu (2012) reported the Young's moduli of four types of pocket coil mattresses. In some studies, mattress samples were only vaguely described. López-Torres et al. (2008) selected four mattresses claimed to cover the full range of firmness, while Low et al. (2017) commented that the   Denninger, Martel & Rancourt (2011) packed cells of cubes with hollow ellipsoidal cavities and customized the dimensions and cavities of these cubes based on individuals' spinal curvature and regional body weight portion. Similarly, Chen et al. (2014) fitted the mattress intermediate layer with the body shape contour, and Park, Kim & Kim (2009) allowed the subjects to adjust the mattress regional height according to their own preferences. Some studies customized mattress regional stiffness to achieve desirable spinal curvature. For example, Leilnahari et al. (2011) andVerhaert et al. (2012a). In addition, Zhong et al. (2014) divided the mattress into five zones and evaluated 14 combinations using six types of spring stiffness in different zones.
Five of the selected articles employed the same active-control mattress system, Dynasleep (Custom8, Leuven, Belgium) (Deun et al., 2012;Verhaert et al., 2011bVerhaert et al., , 2012aVerhaert et al., , 2012bVerhaert et al., , 2013, which was equipped with indentation sensors and adaptive actuator spring pockets, with each mattress cell containing two pocket springs with different Nine females classified into three groups (n = 3) based on BMI Diagnosed musculoskeletal pathology Supine A total of 14 mattresses formed by the different combination of regional stiffness in five zones using six types of spring stiffness. The mattress consisted of a superficial layer of PU foam and a core layer composed of rows of pocketed springs.

NA
Notes: M, male; F, female; BMI, body-mass index; PU, polyurethane; NA, not applicable; NS, not specified. * The authors used the unit of kPa to describe the density of the sleeping support without justification.   Validation of somatotype models demonstrated a high correlation index compared to real data: more than 85% in height and body circumferences; 89.9% in weight; 80.4% in body mass index; and more than 70% in morphotype categorization.

Lee et al. (2015)
Mixed factorial design (gender, body regions, duration), nonrandomized controlled trial Analysis of body pressure and perceived level of pain for different genders, body regions, and durations of supine lying.
Head regions experienced significantly higher pain scores and pressure intensities; lower back was not too high in pressure intensity but featured the second highest pain score; the back and pelvic girdle showed a significant difference between males and females on the pain score; pain appeared in all body regions after 10 min and progressed as time increased.

Lee et al. (2016)
Repeated measurements, nonrandomized controlled trial Comparison of body pressure and perceived level of pain between the floor and mattress for different durations of supine lying.
Head regions featured a significantly higher pressure intensity; the pain scores of all body regions except for legs were significantly higher for the floor condition; the pain score of the floor condition significantly increased at 1 min compared to those of the mattress group.
(Continued ) Design process, repeated measurements, nonrandomized controlled trial Design of a customized mattress with different zonal elasticity that can achieve optimal spinal alignment; Comparison of spinal alignment achieved by firm, soft, and custom mattresses.
The customised mattress with different zonal elasticity afforded better spinal alignment (least π-P 8 ), followed by firm and soft mattresses.
Non-randomized controlled trial, correlation Comparison of perceived firmness, usability, and comfort between young and elderly people; Investigation of the correlation between subjective ratings and results of objective measurements (pressure distribution and objective firmness).
No perception differences between the young and the elderly were found. Significant correlations were found between increments in objective firmness and perceived firmness (positive); increments in average pressure and perceived firmness (positive); increments in objective firmness and average pressure were associated with increments in overall comfort and reductions in rolling difficulty.

Low et al. (2017)
Randomized cross-over, singleblind controlled trial Comparison of the body contact pressure profile of different mattresses in three different postures.
Compared to the case of a PU mattress, reduced peak pressure and a more even pressure distribution was observed for a latex mattress. Subjects preferred height adjustment in W-shape in supine and lateral postures, and in U-shape in lateral prone postures; The adjusted height was significantly correlated with (a) the subjective rating and (b) the ratio of bed sector regional pressure and the total bed pressure.

Verhaert et al. (2011)
Repeated measurements, nonrandomized controlled trial Investigation of the effect of an active-control bedding system autonomously altering stiffness distribution according to the estimated spinal alignment and comparison to a sagging bedding system. The sagging sleep system negatively affected sleep quality in prone and lateral postures; The relationship between mattress design and sleep quality was affected by anthropometry and posture. Estimation showed good correspondence (85%) in comparison to the validated spine shape in terms of score ranking.

Verhaert et al. (2013)
Mattress design process, randomized crossover singleblind controlled trial Presentation of an activecontrol mattress system that can: 1. detect body movement and recognize sleep posture; 2. estimate the shape of the spine by combining indentation with human models; 3. based on indentation measurement and feedback, control the mattress system to achieve optimal spinal alignment by customizing regional mattress stiffness. Performance comparison of the active and non-active modes of the active-control mattress.
The use of the active-control mattress system significantly improved the perceived sleep quality.

Instrument design, repeated measurements
Development of a mattress evaluation method based on body pressure distribution and comparison of back surface and spinal alignment between supine lying and upright standing through finite element simulation. Comparison of the outcomes obtained for a palm fiber mattress and a bilayer latex/ palm fiber mattress.
A novel parameter was proposed by comparing the back surface contours of supine lying and natural standing postures via similarity analysis. The bilayer latex/palm fiber mattress produced a back surface contour close to that of upright standing, which indicated a preferable selection.
intrinsic stiffness coefficients (k = 0.2 and 0.076 N/mm) arranged in a parallel way. Real-time adjustment of regional stiffness was facilitated by the vertical displacement of spring pockets according to sensor-supplied data and the pre-set algorithm. The articles did not disclose the detailed algorithm of the active control design, although the relevant content could be found in the dissertation of one of the authors (Verhaert, 2011). Esquirol Caussa et al. (2017) and Palmero et al. (2017) developed a recommendation matrix for suggesting mattress-pillow-topper combinations. Although the interactions of these components were not studied, the inclusion of different support components is appreciated, since other studies seldom specify the use of pillows and toppers. Importantly, pillow design has a compelling effect on the biomechanics of sleep support, particularly on the cervical spine posture (Ren et al., 2016).

Posture
Supine and side lying postures are frequently evaluated in literature. Most of such studies are controlled trials in which the posture is instructed and maintained during the required time period, for example, for supine posture, subjects were asked to put their hands straight on both sides. In the work of Esquirol Caussa et al. (2017), subjects were instructed to put their feet together, while in the work of Palmero et al. (2017), legs were placed slightly apart. For lateral posture, subjects were normally placed with the body perpendicular to the ground surface (Denninger, Martel & Rancourt, 2011;Leilnahari et al., 2011), which was challenging, since people tend to turn their shoulder forward toward the mattress (Verhaert et al., 2013). Additionally, limb placement was subject to some variations. In some studies, limb flexion was controlled at a given angle (Leilnahari et al., 2011), while in others, subjects were allowed to bend the limb slightly and naturally (Denninger, Martel & Rancourt, 2011). The variation of sleep posture may produce different trunk bending angles and thus may influence the outcome of the Investigation of the relationship between the outcome of computer simulation (finite element analysis) and subjective ratings on preference and comfort.
The subjective ratings corresponded to the prediction outcome, including the von Mises stress of the cervical vertebral region and the sinking displacement of the neck region.

Zhong et al. (2014)
Instrument design, validation (error analysis), mattress design process Estimation of spinal curvature with mattress indentation; Determination of an optimal mattress zonal stiffness.
The overall mean absolute error and mean relative error between the estimation and experimental measurements equaled 3.4 mm (SD: 2.7) and 9.27%, respectively. CTh, ThL, LS generally increased with lower back and hip zone stiffening; the upper body became more levelled with stiffened hip zones and more inclined with stiffened upper back zones.
spinal alignment. The influence of the posture variation requires further investigation. Prone posture was relatively less evaluated and specified (Low et al., 2017). Instead of adopting an experimental approach, some studies simulated and estimated postures using computational models. By scaling a simplified model with subject body dimensions, Verhaert et al. (2012aVerhaert et al. ( , 2012b attempted to evaluate supine, lateral, and prone lying postures on ordinary, customized, and active-control mattresses. Zhong et al. (2014) conducted a finite element analysis using a supine lying human model to assess the influence of mattress regional stiffness variability.
The use of posture detection and prediction algorithms allowed the conduction of observational studies with uncontrolled sleeping postures. Verhaert et al. (2011b) used cluster analysis to categorize the population into two groups according to the time spent in lateral, prone, and supine postures, while Deun et al. (2012) and Verhaert et al. (2013) identified four main sleeping postures (including left-and right-side lying, prone lying, and supine lying) employing indentation data provided by the active-control bed. The above categorization was achieved using a support vector machine according to the combination of indentation, shoulder-hip ratio, knee-hip ratio, total indentation, lateral asymmetry index, and lower leg indentation data, which has been proven to achieve sleeping posture recognition with >90% accuracy (Verhaert et al., 2011a). However, this technique was not sufficiently sensitive to detect intermediate postures, which account for 10% of the sleeping time (Verhaert et al., 2013).

Determinants and evaluation methods
Although body-mattress contact pressure is commonly used to represent the entity of pain and discomfort, the validity of this approach remains controversial (Buckle & Fernandes, 1998). Force or pressure induces the deformation of skin and thus triggers the sensation of touch (via mechanoreceptors) and pain (via nociceptors) upon high loading (Kilinc-Balci, 2011). High pressure can also adversely affect peripheral blood circulation and lead to numbness and discomfort (López-Torres et al., 2008). Therefore, the perception of pain or comfort is believed to be strongly related to the perception of pressure (López-Torres et al., 2008). The body pressure measurement system was characterized by thin and flexible sheet sensors that only minimally interfered with the mattress support (Chen et al., 2014;Lee et al., 2015Lee et al., , 2016Low et al., 2017). Yet, the sensors may disperse the concentrated pressure and thus underestimate the peak pressure. Similarly to the principle of contact pressure, some studies investigated the effects of the regional supporting load using a matrix of load cells (Denninger, Martel & Rancourt, 2011) or an indentation bar (Zhong et al., 2014) and indentation sensors embedded in the mattress (Dynasleep). Peak pressure, average pressure, and contact area were often measured in different body regions with the aim of pressure reduction.
Spinal alignment was the second frequently investigated parameter, since the adoption of neutral or physiological spine curvature is thought to avoid musculoskeletal problems or pain. In fact, spinal alignment or curvature in side-lying postures was frequently evaluated because of the ease of measurement. Several studies limited the measurement of alignment to two dimensions at the coronal plane using a camera (Denninger, Martel & Rancourt, 2011;Leilnahari et al., 2011), while some attempted to perform three-dimensional (3D) measurements using pen-tip optical tracking (Denninger, Martel & Rancourt, 2011), rasterstereography (Verhaert et al., 2011b), a camera equipped with a depth sensor/infrared projector (Kinect, Microsoft, Redmond, WA, USA) (Esquirol Caussa et al., 2017;Palmero et al., 2017), and registration of images from sagittal and coronal planes (Deun et al., 2012;Verhaert et al., 2011bVerhaert et al., , 2012aVerhaert et al., , 2012b. Zhong et al. (2014) approximated spinal curvature in the supine position using a custom-made indentation bar embedded in the mattress, while others attempted to model the supine spine by integration of indentation measurements and computer modeling. Human models were personalized by scaling a generic model with measured body dimensions or 3D scanning (Deun et al., 2012;Leilnahari et al., 2011;Verhaert et al., 2011bVerhaert et al., , 2012aVerhaert et al., , 2012bVerhaert et al., , 2013Wu, Yuan & Li, 2018). Instead of using a simple soft-tissue-lump model, Verhaert et al. (2013) combined the body surface model with a simplified skeleton model to enhance the accuracy of posture approximation, and further improvement was achieved by Leilnahari et al. (2011) through the use of a musculoskeletal model (BRG.LifeMod) accounting for joint stiffness and the range of motion.
There was no consensus on the use of a specific parameter to quantify spinal alignment. For example, a simple approach was used in two cases to identify the locations of each vertebra center (Denninger, Martel & Rancourt, 2011;Leilnahari et al., 2011), while in other cases, the thoracic-lumbar angle was estimated using the regression lines of thoracic and lumbar regions (Leilnahari et al., 2011;Verhaert et al., 2011bVerhaert et al., , 2012aVerhaert et al., , 2012b, since a discontinuity was often observed at the transition from the flexible lumbar to the rigid thoracic regions (Leilnahari et al., 2011). Some angles were also calculated by comparing lines joining the upper and lower regions with the horizontal line (Verhaert et al., 2011b(Verhaert et al., , 2012b(Verhaert et al., , 2013Zhong et al., 2014). Mean distances were measured between the regressed curved line and the horizontal line, the line of the spinal axis, or a reference curve (Verhaert et al., 2011b(Verhaert et al., , 2013Zhong et al., 2014), and root-mean-square deviations were computed to quantify deflections from the desired curvature. However, no study investigated the twisting of the spine or trunk segments, which can be overlooked in some occasions of back pain.
Finite element analysis provides a versatile platform to predict the internal biomechanics of the body in a controlled environment (Wong et al., 2014). Regarding internal stress and strain, Yoshida, Kamijo & Shimizu (2012) performed finite element analysis to examine the von Mises stress of the second to fifth cervical vertebra for different mattress firmness, additionally comparing sinking displacements in head and thoracic regions. Wu, Yuan & Li (2018) modeled the back contour of the human body by finite element simulation and compared it with that observed during natural standing. In addition, Denninger, Martel & Rancourt (2011) constructed a simplified finite element model of the whole body and optimized the design of mattress cells by equalizing the body portion weight with the supporting force of each mattress cell. The process was performed assuming a minimal trunk shear force predicted by finite element analysis. The limitations of finite element simulation include model simplifications and assumptions on pre-defined sets of loading cases (Wong et al., 2017). It remains difficult to reconstruct a few anatomically detailed models with corresponding experiments for validation (Wong et al., 2018). Oversimplified models could have problems on the prediction accuracy and validity that could limit practical applications (Wang, Wong & Zhang, 2016). Non-biomechanical methods were implemented to correlate biomechanical parameters with sleeping quality and subjective feedback on comfort using polysomnography and questionnaires. The details of these methods are beyond the scope of this review and can be found elsewhere (Radwan et al., 2015).

Optimization or selection criteria
Since contact pressure and spinal curvature/alignment are the predominant parameters of interest, it is important to know the desirable range or values of these parameters to suggest the directions of design optimization and realize high-quality mattresses.
Mattress design often strives for lower contact pressure in view of the fact that high pressure may cause discomfort and sore formation (Esquirol Caussa et al., 2017;López-Torres et al., 2008;Low et al., 2017;Palmero et al., 2017). Low et al. (2017) aimed to reduce peak pressure and realize a more even pressure distribution. Additionally, Esquirol Caussa et al. (2017) and Palmero et al. (2017) concluded that a soft topper should be implemented when more than three points with pressure exceeding 60 mmHg are present, while a medium-density pillow should be used when the maximum pressure at the occipital region falls between 30 and 40 mmHg. Conversely, several authors opposed this view by showing that high pressure and discomfort should not necessarily be correlated (Lahm & Iaizzo, 2002;Lee et al., 2016). Chen et al. (2014) commented that to achieve better sleep quality, the body pressure distribution should neither be over-concentrated nor over-distributed. Different body regions can exhibit different pressure tolerabilities (Lee et al., 2015), while the comfort of mattresses with different stiffness can be perceived differently depending on body built or body weight (Yoshida, Kamijo & Shimizu, 2012).
A straight horizontal line in the frontal plane was employed for evaluating spinal alignment in a lateral lying posture. A scoliotic spine position was regarded as non-natural or non-physiological and was believed to result in muscle imbalance and back pain (Aebi, 2005). The S-shaped curvature of the spine in the sagittal plane was of particular interest for supine lying. Denninger, Martel & Rancourt (2011), Wu, Yuan & Li (2018), and Zhong et al. (2014 assumed upright spine alignment or curvature as the desired alignment, while Verhaert et al. (2013) stated that the targeted upright spine should consider a slightly flattened lumbar lordosis to accommodate the switched working axis of gravity. In addition to spine alignment, Wu, Yuan & Li (2018) compared the back surface contour obtained for simulated supine lying with that determined by 3D scanning during natural standing and proposed a comfort index based on similarity analysis.
Compromising and weighing of two or more determinants to establish a single measure remains difficult. Verhaert et al. (2012a) formulated an ergonomic bed score (EBS_L) based on a weighted combination of lumbar and thoracic angles, whilst Denninger, Martel & Rancourt (2011) presented an expert system that considered both spinal curvature and trunk shear during the design process. However, neither of these authors considered the trade-off between such criteria and objective functions. The study of Wu, Yuan & Li (2018) was the only one that considered multiple dimensions such as body pressure, back surface contour, and spine alignment. The obtained results showed that these parameters provided conflicting conclusions toward better mattress construction, and it was decided to use the back surface contour as the determinant after alleged comprehensive consideration.

Study design and key findings
The reviewed articles had diverse study objectives and hence, study designs, as shown in Table 3. Five articles investigated the processes and methods of designing new mattresses, including those with customized regional stiffness and height (Denninger, Martel & Rancourt, 2011;Leilnahari et al., 2011;Park, Kim & Kim, 2009;Verhaert et al., 2013;Zhong et al., 2014). Custom-made mattress that was constructed with different zonal elasticities produced the smallest thoracolumbar angle (4.10 ) compared to the firm (8.9 ) and soft surface (12.66 ) (Leilnahari et al., 2011). Zhong et al. (2014 suggested that a custom-made mattress with stiffening of the lower back and hip regions would increase the cervicothoracic, thoracolumbar, and lumbosacral angles, while the stiffening of the upper back region would decrease these angles. Park, Kim & Kim (2009) allowed the subjects to adjust the regional heights of the mattress and discovered that there was a significant correlation between the regional pressure ratio differences and subjective ratings. They preferred the W-shaped bed in both supine and side postures and U-shaped bed in prone posture, compared to the flat bed (Park, Kim & Kim, 2009).
Using different objective functions, three research teams developed recommendation models for the optimal selection of mattresses or combinations of sleep system constituents (pillow, mattress, and topper) (Esquirol Caussa et al., 2017;Palmero et al., 2017;Verhaert et al., 2012a). Another three studies involved the design of instruments or techniques to estimate spinal curvature during supine lying, which is otherwise difficult to assess because of the lack of back exposure (Verhaert et al., 2012a(Verhaert et al., , 2012bZhong et al., 2014).
The introduction of new instruments and the application of computer simulations call for a validation process. Percentage errors or deviation values were common and simple parameters used for validation (Denninger, Martel & Rancourt, 2011;Verhaert et al., 2012a;Zhong et al., 2014), while some studies conducted correlation analysis and introduced confusion matrices (Esquirol Caussa et al., 2017;Palmero et al., 2017;Verhaert et al., 2012b).
Comparative studies were conducted to evaluate conventional and newly designed mattresses. Chen et al. (2014) and Low et al. (2017) implemented a randomized cross-over single-blind controlled trial to evaluate different mattress materials. Low et al. (2017) found that latex mattresses can significantly reduce peak pressure by up to 35.1% compared to that of the polyurethane mattresses. Besides, mattresses with over-concentrated or over-even pressure distribution produced low satisfaction scores and were proven not beneficial to sleep quality (Chen et al., 2014). Deun et al. (2012) and Verhaert et al. (2011bVerhaert et al. ( , 2013 evaluated an active-control bed by comparing it to that with a non-active mode or an exaggerated sagging condition by repeated measurements. Their results demonstrated that the active-control bed significantly improved subjective ratings (sleep quality, daytime quality, perceived number of awakenings), polysomnographic measurements and spinal alignment.

Remarks
Herein, we reviewed the state-of-the-art biomechanical research on sleeping mattress design, particularly its scope and methodology, demonstrating that mattress research and development have shifted from homogeneous material evaluation to regional characteristic customization. Several authors attempted to adjust the height or stiffness of different zones to facilitate the adoption of physiological spine curvature. An active-control bed system was shown to enable regional stiffness change in real time to accommodate different postures, and several studies used the same active-control bed system but employed slightly different control algorithms for a better response. In fact, a market research conducted by TechNavio identified smart mattresses with sleep tracking, movement detection, and automatic firmness adjustment functions as the next major market driver in the mattress production industry.
To date, pressure is one of the gold standards for mattress performance evaluation, despite providing results that are in subtle conflict with those obtained using sufficient supporting force as an evaluation parameter. For instance, low pressure can be used to improve tactile comfort at the cost of a sagging spine and sinking lumbar region, while the introduction of expensive support surfaces to eliminate pressure is not necessarily beneficial (Goossens, 2009;Lahm & Iaizzo, 2002). Esquirol Caussa et al. (2017) and Palmero et al. (2017) aimed at a pressure threshold of 30-40 mmHg, with attenuation performed by varying the density of pillows and toppers. The threshold value was defined according to the premise that subcutaneous ischemia happens at a critical capillary close pressure of >30 mmHg (McCall, Boggs & Letton, 2012). However, it should be noted that this assumption of equating capillary pressure and skin contact pressure may not be valid for the trunk, which features thick soft tissues and muscles. In fact, in relevant studies, measurement were performed at the fingernail fold (De Graaff et al., 2002). Lee et al. (2015Lee et al. ( , 2016 demonstrated that different body regions feature different pressure sensitivity and tolerability, highlighting the need to establish a reference quantitative range of acceptable pressure at different body regions for mattress design. Spinal alignment is believed to reflect the complex biomechanical interaction between the human body and the sleeping support (Verhaert et al., 2012a). However, the measurement of spinal alignment in the supine position remains challenging because of the lack of back exposure and the fact that instrument placement may interfere with body support. Although such estimations are commonly performed by computer models referenced to the results of 3D scanning during upright standing, we proposed the use of optical Fiber Bragg Grating (FBG) sensors for an unobtrusive assessment of spine alignment (Sadek et al., 2017). In particular, the suggested flexible thin wire FBG sensor allowed real-time 3D geometry sensing without causing much interference to the support surface (Allsop et al., 2012;Ryu & Dupont, 2014) and thus enabled the detection of different planes including body twisting, which was often overlooked. Moreover, the FBG system can also sense temperature and humidity simultaneously, which are important attributes affecting sleeping comfort (Zhang et al., 2010).
Similarly, the objective functions or criteria of spine alignment in supine postures are questionable, although Verhaert et al. (2012a) viewed it as the primary metric of the overall body deformation. Moreover, whereas upright standing can be the most available posture to be referenced, it is theoretically inappropriate to regard upright standing as a desirable alignment for supine sleeping, since the spine loading modes during standing and lying are totally different. Under the influence of gravity, the curvature of the standing spine tends to be vertically exaggerated because of the induced compression, while that of the lying spine tends to be flattened (Verhaert et al., 2012a). The role of intervertebral discs, muscles, and ligaments is expected to be more prominent during standing, which affects spine curvature and the perception of relaxation and comfort. To this end, we have discovered that the spine alignment desirable for supine sleeping can be determined under free-floating conditions. During flotation therapy (flotation spa), participants lie in a tank filled with salt-saturated water in a sound-, light-, and temperature-controlled environment (Van Dierendonck & Te Nijenhuis, 2005). The high density of the saline is believed to provide a sufficient and appropriate buoyancy force to support the body and has been proven to bring about relaxation responses, including the sense of zero-gravity and reduced muscle tension (Van Dierendonck & Te Nijenhuis, 2005). We believe that this approach could be a possible alternative to establishing a desirable supine lying posture, whilst the water tank can also allow exposure for measurement.
Despite the fact that most studies attempted to ease discomfort only, comfort is not exactly the opposite of discomfort (Goossens, 2009). Helander (2003 defined comfort as a sense of relaxation and relief. Likewise, Verhaert et al. (2012b) argued that comfortable bedding should facilitate the relaxation of muscles and intervertebral discs to recover from day-long loading. The lack of skeletal support was also regarded as a consequence of prolonged muscle activation that triggered a guarding action from the brain (Russell, 1999). However, it remains difficult to measure muscle loading and the deformation of intervertebral discs or other soft tissues during sleep. While computer simulations used in the reviewed articles employed simplified models to estimate spinal alignment, anatomically detailed finite element models and musculoskeletal models accounting for the functions of muscles and the geometry of the intervertebral discs can provide more complementary evidence on the objective representation of mattress comfort.
Mattress design and selection to achieve client satisfaction are recognized as a tedious trial-and-error process. This broad review was conducted to systematically sample the relevant information and thus improve the process of mattress design and selection as well as stimulate pertinent standard setup. While the studies implemented personalized design, cutting edge technology and algorithm to improve mattress design, the basic requirement of mattress design is inconclusive. It is pragmatically demanding to evaluate the influence of mattress material (stiffness), thickness, shape, and pillow combinations on the pressure distribution and spinal alignment systematically. This study features several limitations. First, we did not assess methodological quality, since the involved designs and scopes were diverse and thus difficult to compare. Second, a number of works dealt with anti-pressure-sore mattresses and were thus beyond the scope of this review. Moreover, eligible articles were determined by several manual screening processes and discussion, and the repeatability of the search and screening may thus be challenged.

CONCLUSIONS
Future studies should aim to establish and justify reference values for mattress design and selection as well as develop an algorithm for determining the trade-off between the weighting of different determinants.

ADDITIONAL INFORMATION AND DECLARATIONS Funding
The work was supported by the Shenzhen Foundation Research Fund granted by the Shenzhen Science and Technology Innovation Committee (JCYJ-20160531-1846-21718), and the NSFC granted by the National Natural Science Foundation of China (11732015). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.