A review of portable quantitative and semi-quantitative devices for measurement of vitamin A in biological samples

Graphical abstract Assessment of vitamin A concentrations in biological specimens or functional indicators of vitamin A status (severity of night blindness or lack thereof) require field-friendly methods for assessment to expand coverage, particularly in low-income settings. In this review, portable devices for measurement were compared with reference methods across key criteria. For example, portable devices to determine vitamin A concentrations in blood or milk samples were catalogued and compared with the reference method of high performance liquid chromatography, and a portable dark adaptometer to assess vision was compared against the reference of a Goldmann-Weekers dark adaptometer.


Introduction/background
Vitamin A deficiency (VAD) continues to be a major global health issue leading to poor health outcomes, including night blindness, greater severity of measles infection, and higher mortality risk from infectious diseases (Tanumihardjo et al., 2016). Most existing analytical techniques to assess vitamin A status by measuring serum retinol or retinol binding protein require access to a sophisticated laboratory and equipment such as high performance liquid chromatography (HPLC) (de Pee and Dary, 2002). These methods require extensive sample preparation, are time-consuming, and are potentially prohibitively expensive, depending on the number of samples to be analyzed. Furthermore, VAD is more prevalent in lower income countries, where such laboratory resources may be limited or might not yet exist; in recent vitamin A surveys, <20% of pregnant women at risk have been covered by population surveys globally, possibly partly because of a lack of diagnostics (WHO, 2009).
Portable, field-friendly devices and tools for assessing vitamin A status in populations have the potential to overcome some of the limitations of traditional, laboratory-based testing. These methods may differ in their cost, accuracy, reliability, ease of use, and required consumables/reagents for performing the testing.
A review cataloguing the range of portable tests for vitamin A status and VAD in biological samples, and summarizing these devices' performance with respect to a reference standard method, is not available. Therefore, the goal of this review was to enable current manufacturers to modify and improve their products according to the gaps identified herein, and to set design goals for new products meeting the current demands of industry, regulators, and other stakeholders.

Materials and methods
In December 2020, we conducted a standardized search of the literature indexed in five databases (MEDLINE, EMBASE, World Health Organization Global Index Medicus, Scopus, and Web of Science) with no restrictions on language, location, or date of publication. We designed a search strategy for MEDLINE (PubMed) (Supplementary Table 1) and translated the search strategy for the remaining databases with guidance from the evidence synthesis specialists at Mann Library, Cornell University. We also used an online search engine to search for other sources such as manufacturers' websites and patents, and we consulted with subject matter experts within our organizations to gain more information.
We catalogued any portable devices measuring vitamin A or vitamin A deficiency in biological samples, either as reported in studies or provided on manufacturers' websites. We included both portable devices/methods measuring vitamin A status and devices/methods that indicated VAD. Initially, we considered devices measuring skin carotenoids, as shown in our search strategy; however, because of the lack of established guidance or consensus regarding the conversion of skin carotenoid measurements via Raman resonance spectroscopy (e.g., BioPhotonic Scanner (Pharmanex/Nuskin Enterprises, 2018)) to blood carotenoid measurements and overall vitamin A status (von Lintig, 2020), we determined that these devices were beyond the scope of the review.
The inclusion criteria for our analysis of device performance included certain study designs such as proof-of-concept development studies, method comparison studies, and diagnostic test accuracy studies; studies involving human participants (e.g., observational studies or randomized controlled trials) were considered if the authors described using a portable method for analyzing vitamin A in biological samples. Animal studies were also included. Eligible studies were required to measure vitamin A in any biological sample, including blood, eyes, or breast milk, with a portable device and to compare the device performance with that of a reference method, such as HPLC, depending on the sample type. Studies detailing field friendly methods of sample collection (e.g., dried blood spots) necessitating the use of a non-portable device or a laboratory for analysis were considered beyond the scope of this review.
We contacted the authors to request raw data or more information as needed. We also re-analyzed raw data, when available, as needed.

Results and discussion
Catalog of portable devices From our search ( Fig. 1), we catalogued 25 portable devices, kits, and/or field-friendly assays able to assess a variety of biological sample and vitamin A biomarker types in Table 1a (blood, milk) and Table 1b (eyes/vision assessment).

Vitamin A deficiency biomarkers
In Table 2, we list definitions of VAD used across studies for a variety of biological sample types, from humans or cattle, including cows, calves, and bulls (Table 2). We also note which studies used particular definitions (e.g., VAD measured as RBP ≤ 0.70 µmol/L was measured by Hix et al. 2004). A previous review by Tanumihardjo (2016) has outlined the utility of biomarkers for vitamin A nutrition status (Tanumihardjo et al., 2016), which we adapted for Table 2. We outline the biomarkers of vitamin A as identified in our literature search below.
Vitamin A liver concentration (µmol vitamin A/g liver) is the gold standard for vitamin A status but requires invasive techniques such as biopsy to be measured (Tanumihardjo et al., 2016). Sampling blood enables the quantification of serum retinol, serum retinol-binding protein (RBP), or provitamin A in the form of beta-carotene; however, each measure has trade-offs. Serum retinol reflects liver stores only at extremes of deficiency (≤0.07 µmol/g liver) or elevation (>1.05 µmol/g liver) (WHO, 2011), because serum retinol is homeo-statically regulated by the body. The World Health Organization defines VAD as serum retinol ≤ 0.70 µmol/L (WHO, 2011). RBP is commonly assumed to have a 1:1 ratio with serum retinol, and therefore the same cut-offs are sometimes used for both retinol and RBP. However, this ratio can be affected by the extent of VAD, zinc deficiency, acute phase response, protein-energy malnutrition, liver disease, acutely stressful situations, high fever, antibiotic use, or obesity (Tanumihardjo et al., 2016, de Pee andDary, 2002). Therefore, previous studies have proposed other deficiency cut-offs, such as 0.69 µmol/L (Semba et al., 2002) or 0.83 µmol/L (Engle-Stone et al., 2011, Gorstein et al., 2008. Recently, the Global Alliance for Vitamin A has recommended analysis of a subsample by HPLC to confirm the cut-off point for VAD; furthermore, given the acute phase response, inflammation markers such as C-reactive protein and alpha-1-acidglycoprotein must also be measured (Global Alliance for Vitamin A, 2019).
Beta carotene is one of several dietary provitamin A carotenoids, a plant-derived form of vitamin A. The body converts dietary provitamin A carotenoids into retinol with the following conversion factors: 1 µg retinol activity equivalent (RAE) equals 1 retinol equivalent (RE), 1 µg retinol, 2 µg β-carotene in oil, 12 µg β-carotene in mixed foods, or 24 (12-26) µg other provitamin A carotenoids in mixed foods (Institute of medicine, 2001, Combs and McClung, 2017, Blaner, 2020. The conversion efficiency ratio of beta carotene to RAE is still debated. For example, the European Food Safety Authority suggests that the conversion is 6:1 rather than 12:1 (EFSA Panel on Dietetic Products, Nutrition Allergies, 2015). Carotenoids can be measured in blood, milk, or skin, and several studies have found a positive association of skin carotenoid concentrations with serum or plasma carotenoid status (Zidichouski et al., 2009, Aguilar et al., 2014, Morgan et al., 2019, Hayashi et al., 2020. However, a consensus has not been reached regarding a conversion factor or how the measurements equate to vitamin A status (von Lintig, 2020). Because carotenoids tend to reflect recent dietary intake rather than long-term status, recommended serum carotenoid deficiency cut-offs have not been estab-lished in humans (von Lintig, 2020). Deficiency in β-carotene in cow's blood has been defined as 0.6-1.5 mg/L (Klein et al., 2013, De Ondarza and Al, 2009, Schweigert and Immig, 2007. In breast milk, retinol may be measured to estimate both the maternal vitamin A status and intake, and the infant intake of vitamin A (Engle-Stone et al., 2014, Tanumihardjo et al., 2016. Additionally, breast milk retinol measurement is influenced by the stage of lactation, time of day, "fullness" of the breast, feeding status if milk from both breasts is analyzed, and whether the milk is hindmilk compared with foremilk (Tanumihardjo et al., 2016). VAD is defined as a milk retinol concentration ≤ 1.05 µmol/L, or ≤ 8 µg/g milk fat (Tanumihardjo et al., 2016). In cows, milk β-carotene levels are often measured and linked to bovine fertility and health.
Because of vitamin A's role in in producing rhodopsin, the visual pigment of rods in the eyes, VAD can cause ocular manifestations resulting in poor vision (World Health, 2014). These include night blindness, conjunctival xerosis, Bitot's spots, corneal xerosis, and keratomalacia. Impaired adaption to the dark is among the first symptoms of VAD, and it can be used as a screening tool (World Health, 2014). Tests such as pupillary and visual thresholds can assess dark adaptation by determining the lowest-intensity level of light required to cause pupillary dilation or to visualize an image (World Health, 2014.
We also identified many studies that used a portable device for assaying samples but did not compare the results to those of a reference method and instead cited previous validation studies. Although the devices used are catalogued and described (Tables 1a and 1b), these studies are not further detailed in this review.
Study populations were mostly from the US and Germany, in addition to Thailand, Italy, France Ireland, Japan, Ethiopia, Morocco, Cameroon, Papua New Guinea, Nicaragua, Cambodia, and Oman. Portable fluorometers, photometers, enzyme-based assays or immunoassays, microfluidics-based approaches, and a dark adaptometer for eye function were assessed and compared with their respective reference standards. Table 4 compares the stated performance criteria described by the device manufacturers' websites to reporting from individual studies using the devices, according to the WHO Affordable, Sensitive, Specific, User-friendly, Rapid and robust, Equipment-free and Deliverable to end-users (ASSURED) criteria for diagnostic tests in resourcelimited settings (Kosack et al., 2017). Only the iCheck Fluoro and iCheck Carotene stated performance criteria on the BioAnalyt website and are included in this table. No studies described the cost of the devices. Most devices did not have published cost information, aside from some of the slit lamps showing list prices. Other devices described by the studies were developed as proofs of concept and are either not on the market or are on the market, but lacking performance criteria on the manufacturer's website.
We note that the study by Ghaffari et al. (2019) reported both measurements of retinol in whole blood and beta carotene in plasma, but only directly references the iCheck Fluoro device and not the iCheck Carotene (Ghaffari et al., 2019). Whereas the iCheck Carotene requires 0.4 mL of sample, the iCheck Fluoro requires 0.5 mL; the authors stated using only 0.5 mL of sample.
Additionally, the manufacturer (BioAnalyt) lists only "colostrum, cattle whole blood, and serum" as appropriate sample types for the iCheck Carotene; however all studies using the iCheck Carotene, including a BioAnalyt report, also analyzed plasma for beta carotene content (BioAnalyt, 2020; Ghaffari et al., 2019;Raila et al., 2012).
A major gap across all devices is the lack of reporting on sensitivity and specificity compared with a reference standard method.
Most studies compared a portable device to a reference method. Tables 5a-5g show the performance of these devices against their reference standards for measuring vitamin A and VAD. Additional analyses conducted with other index (e.g., index 2) tests are described in the text.
In human blood samples, both the iCheck Fluoro and the CRAFTi portable fluorometers were used to measure retinol (Table 5a). The iCheck Fluoro studies showed a high correlation (0.98) and an Rsquared values over 0.95 with respect to HPLC. Both CRAFTi studies found a mean difference in serum retinol of −0.07 µmol/L, and the 2011 study found moderate sensitivity and specificity in identifying VAD at either ≤ 0.70 µmol/L or ≤ 1.05 µmol/L. Few additional comparative data were available between studies. Bias analysis indicated an acceptable level of agreement (within two SDs or 95% acceptability limits) between these devices' performance and HPLC.

Function
Scotopic sensitivity hand-held illuminator (LKC Technologies, Inc.) (Congdon et al., 1995, Sanchez et al., 1997, Peters et al., 2000 Portable field dark adaptometer (PFDA) or digital pupillometer , Palmer et al., 2016 Normal: ≥-1.24 log cd/m 2 (Congdon and West, 2002) Abnormal: ≥stimulus #9 (Congdon et al., 1995) ≥-0.575 log cd/m 2 (Congdon et al., 1995) i.e.,≥20% ( the report was excluded from our primary results and is not in Table 5a above. Four studies analyzed either human or cow's milk samples with the iCheck Fluoro (Table 5b). The device performance varied: some studies reported lower, equivalent, or higher retinol values than those of HPLC. The R 2 values for the correlation between the device and HPLC ranged between 0.35 and 0.79 after adjustment for milk fat content.
In one study (Schweigert et al., 2011b), the authors tested increasingly diluted cow's milk samples with 3.5% fat by using the iCheck, which showed linearity at an R 2 of > 0.99 between 100 and 2500 µg RE/L. The same study also showed a positive correlation between percentage milk fat and µg RE/L milk with the iCheck Fluoro. Precision was tested over an operational range of 60 to 600 µg RE/L, and the inter-assay CV was < 3.5% (not shown in table). RBP was measured in blood with two field-friendly immunoassays reported across three studies comparing a portable device to a reference method (Table 5c). A rapid enzyme immunoassay (RBP-REI), available from Scimedx Corp, was able to detect serum RBP within a range of 10-40 µg/mL, which correlated with the HPLC results (R 2 = 0.79 to 0.86) (Hix et al., 2004, Hix et al., 2006. The RBP-REI assay was also compared with another portable, laboratory-based device (index 2), a commercially available radial immunodiffusion plate reader (RID; The Binding Site, San Diego), by measuring RBP in 40 serum samples (Hix et al., 2004). Compared with the higher R 2 values in validation against HPLC (R 2 = 0.82 and 0.86; Table 5c), the RBP-REI had a lower, but still acceptable, correlation with the RID method (R 2 = 0.73; linearity: y = 0.50x + 0.45) (not shown in table). Comparison of RID and HPLC indicated a slightly lower correlation (R 2 = 0.71) (Table 5c). Other validity and precision data were not reported for the comparisons of REI vs. RID, or RID vs. HPLC. From the current manufacturer's website (accessed date: March 15, 2021) (The Binding Site, 2020), RBP was not listed among the human proteins for assessment with the RID plate reader.
A semi-quantitative antigen-antibody binding assay allowed for detection of low concentrations of RBP in serum samples (Ciaiolo et al., 2015). However, because only six samples were used for validation, drawing a conclusion regarding the efficacy of this method is difficult.
We identified two microfluidics-based devices, the EE-µPAD (Lee et al., 2016) and the Tidbit with HYPER filtration , Lu et al., 2018, both of which were able to separate whole blood into serum, detect VAD at high sensitivity and specificity with respect to the reference ELISA test, and send results to a mobile device (Table 5d). We note that the ELISA test may not be a suitable reference method for assessing VAD, owing to inherent problems with antibodies to RBP. Neither device is currently on the market.
Although we identified several portable dark adaptometers (Table 1b), we found only one validation study between a portable dark adaptometer, the Scotopic Sensitivity-Tester 1 by LKC Technologies, and a reference standard, the Goldmann-Weekers dark adaptometer used in clinical settings (Peters et al., 2000) (Table 5e). The portable device was comparable to the reference standard in its sensitivity in identifying elevated final thresholds for dark adaptation, with a correlation (R 2 ) of 0.77. However, this study was performed in the US in an eye clinic, and it remains to be tested and compared with the reference standard in field settings.
The iCheck Fluoro, a portable fluorometer, was used to measure bovine blood samples for retinol (Table 5f). Compared with HPLC, mean differences in whole blood, plasma, or serum retinol ranged from −0.01 µmol/L to 26.5 µmol/L, and the iCheck generally displayed higher values than HPLC. The correlation between the iCheck Fluoro and HPLC was positive, ranging in R 2 values from 0.61 to 0.96. Weaker correlations were observed in cows (range: 0.78-88) than calves (0.90-0.96). Raila and et al. (2017) also compared the correlation between bovine whole blood retinol (n = 10) and plasma retinol (n = 10), both measured by the index test iCheck Fluoro, and found a significant positive correlation (R 2 = 0.87) (Raila et al., 2017). No studies reported sensitivity and specificity, or distinguished specific %CVs. Bias analysis indicated acceptable agreement between the device performance and HPLC.
The iCheck Carotene, a portable photometer, was used to measure carotenoids in bovine whole blood and plasma. Mean differences in beta-carotene concentration ranged from −0.29 mg/L to 0.26 mg/L in plasma samples in cows and calves (Table 5g). The correlation between iCheck Carotene and HPLC was high, with R 2 between 0.93  Palmer et al., 2016) Rod function [dark-adapted rod full-field electroretinogram responses (Peters et al., 2000)]

Table 5b
Portable fluorometers: device performance in human and bovine milk samples.

Gaps and recommendations
On the basis of our review of the literature, portable devices fell into five categories: 1. Portable fluorometers 2. Portable photometers 3. Field-friendly immunoassays and/or microfluidics-based devices 4. Slit lamps 5. Dark adaptometers We found that, although many portable devices for quantifying vitamin A have been developed and described, only a few devices appear to be currently on the market or commercially available; of these, only two had easily accessible performance criteria information on the manufacturers' websites related to vitamin A measurement. Studies tended not to report on portable device characteristics.
Some major gaps involve the lack of data reported by studies. Few studies have reported the portable device's sensitivity and specificity in detecting VAD compared with the reference standard method-a necessary metric for validation and adoption by randomized trials. Furthermore, only the iCheck devices were assessed in more than two studies; other devices should be analyzed further for validation.

Minimal set of criteria for point-of-need devices
See Fig. 2. The device should: 1. Be lightweight with a small form factor for easy transport to the necessary location as needed. 2. Be standalone without needing additional equipment and selfpowered, and should pre-store all the required reagents for the test, and use common reagents that are available on the market.   , Lu et al., 2018)  Notes: MD, mean difference; MFR, multi-faceted ratio i.e., the ratio of the light transmission in the test area to that in the background control area, calculated for RBP for each sample repeat. NR, not reported; RE, retinol equivalents defined as the sum of retinol and retinyl esters, equal to 3.3 International Units (IU) of vitamin A or as 1 µg (units reported by manufacturer-however, retinol activity equivalents (RAE) are the preferred unit for reporting (Institute of medicine, 2001); RMSE, root mean squared error; VAD, vitamin A deficiency; VAI, vitamin A insufficiency. a Reference ELISA utilized samples that were filtered using HYPER system. b Units: µg/mL, mg/L, or µmol/L. 3. Be easy to use with minimal processing steps in the protocol, and should require minimal training effort. 4. Have analytical performance (e.g., %CV < 5% or within Bland Altman 95% limits of agreement) comparable to those of the current laboratory standards, with a capability to test various biological samples. 5. Be affordable and capable of scaling up with locally available consumables where needed. 6. Be able to connect to the internet or an external hard drive with a built-in data management system to allow the test results to be reliably stored and transferred. 7. Be able to output test results quickly and present in a format that is easy to interpret.

Conclusions
In this review, we identified 25 portable methods or devices for a variety of biological sample types including those of human (blood, milk, and eye/vision) and animal (blood and milk) origin. These included nine methods measuring biochemical markers of vitamin A or VAD (serum retinol, RBP, milk retinol, retinyl palmitate, and retinyl esters) and 17 portable methods measuring functional biomarkers (measures of eye health, for example dark adaptation).
The iCheck devices, including iCheck Carotene and iCheck Fluoro -for measuring total carotenoids or beta-carotene, or for measuring retinol, retinyl palmitate, retinyl acetate, or other esters, respectively, in blood or milk-were the only devices with manufacturer-reported performance metrics as well as the most information and data available to ascertain the method's accuracy and precision with respect to those of a gold standard such as HPLC. These methods, in addition to the CRAFTi portable fluorometer, as compared with HPLC, were thus considered acceptable for measuring both blood and milk for biochemical biomarkers of vitamin A and detecting vitamin A deficiency. In measuring human or cow milk samples' retinol concentration, the iCheck Fluoro had variable performance across studies, including both lower and higher values than the gold standard HPLC, thus leading to weaker correlation values than those calculated for blood samples. However, the mean differences were<1 µmol/L, and the values were considered to be within the expected variance. Correlation was improved by diluting the samples; dilution may be required for higher accuracy when the portable method is used.
Several portable immunoassays (RBP-REI, RID, general immunoassay) and microfluidics-based methods (EE-µpad, TIDBIT with or without HYPER platform) for measuring RBP in human blood had acceptable correlations with HPLC reference methods and similar detection of VAD. However, these assays appeared not to be commercially available.
One study has measured eye function with a portable dark adaptometer (Scotopic Sensitivity Tester-1), which had comparable results to the gold standard, a Goldmann-Weekers dark adaptometer. However, field studies using this device in comparison to a reference remain to be performed. Given the importance of eye health as a functional indicator of vitamin A deficiency, this gap in the literature is substantial.
Finally, the iCheck Fluoro was used for measuring bovine blood samples for retinol. Generally, the retinol values were higher than those in samples tested by HPLC. Retinol measurements in calves appeared to have stronger correlations than retinol in cow's blood.
Several studies examined the accuracy of the iCheck Carotene, as compared with HPLC, in determining carotenoid content in cow's blood. Strong correlations with acceptable levels of agreement were observed between device performance and HPLC performance.
In summary, the iCheck devices are commercially available and are acceptable for measuring vitamin A in blood and milk, on the basis of the available data. Many of the other identified devices were proofs of concept and not yet commercially available. Several gaps remain, including studies comparing the other portable devices against a gold

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.