Qualitative discrimination of low-dose oral contraceptives and identification of environmentally stressed tablets: Comparing a handheld near infrared spectrometer to a benchtop spectrometer

Abstract

Levonorgestrel – ethinyl estradiol tablets (Levo-EE) are an essential medication provided through global health supply chains for family planning purposes. Ensuring the quality of Levo-EE and other essential medications in a global supply chain is a primary concern. Portable and handheld diffuse reflectance spectrometers have become more approachable from a cost and usability perspective in recent years. A discriminatory screening method with a handheld near infrared (NIR) spectrometer (900 – 1700 nm) differentiating between brands of Levo-EE and placebos has been constructed. Additionally, the handheld spectrometer was used to determine if environmentally stressed tablets could be identified and differentiated in a qualitative screening method. The same samples were scanned on a benchtop diffuse reflectance NIR spectrometer as well (350 – 2500 nm) for comparison. Brands of oral contraceptives were able to be discriminated by applying a Mahalanobis distance-based classification approach to the handheld spectrometer, though be it at a slightly lower level of discernability due to the reduced spectral range collected. Environmentally stressed samples were applied to the classification model, which were flagged as unlike the reference dataset. Spectra from the environmentally stressed tablets saw an increase in absorbance for water associated peaks with the benchtop spectrometer at approximately 1450 nm and 1949 nm for tablets stored at 30, 40, and 50°C–54°C. Data collected by the handheld showed that the spectrometer offers a low-cost screening approach to oral contraceptive tablets containing two low-dose active ingredients. The handheld spectrometer was also able to flag environmentally stressed samples. The total model accuracy for the target product (brand “A”; N = 560) was above 98% for both benchtop and handheld spectrometers. While these results focus on oral contraceptives, a similar approach could be conducted for other solid dosage tablets for more approachable quality compliance screening.

Keywords

Portable spectroscopy levonorgestrel ethinyl estradiol Mahalanobis distance environmental stress

Introduction

Oral contraceptives such as levonorgestrel – ethinyl estradiol (Levo-EE) are deemed essential medications by the World Health Organization (WHO).¹ Oral contraceptives are a common form of family planning and supplied through global health initiatives.² There are approximately 1.9 billion women of reproductive age globally, and an estimated 842 million are using contraceptive methods.³ Clinicians commonly prescribe two low-dose active pharmaceutical indigents (API)s with one being a progesterone and the other an estrogen, which may offer other advantages in preventing venous events such as blood clots.⁴ By the year 2030, the number of age bearing women will increase, especially in sub-Saharan Africa.³

Poor-quality and falsified pharmaceuticals are an ongoing problem in supplying essential medications to those in resource limited areas including medications that are substandard, falsified, or degraded.^5,6 These substandard medications risk prolonging illness, failed treatment, and possibly mortality. A previous meta-analysis found the prevalence of substandard medications reported in the literature continues to rise, however, estimating an economic impact was difficult and could range between $10 and $200 billion USD annually.⁷ Rapid diagnostic tools for substandard medications are necessary to the response and key to quality compliance.⁸

Diffuse reflectance spectroscopy (DRS) methods have been used for a number of years in pharmaceutical quality compliance of incoming raw materials (Igne and Ciurczak, 2021a),⁹ and occasionally to monitor finished pharmaceutical products compliance.^10,11 DRS methods use light interactions with tablets and suspensions to capture information pertaining to the physical and chemical nature of the samples.¹² In terms of a solid dosage tablet, this can relay information relating to the API and excipients.¹³

In a global pharmaceutical supply chain, careful attention to detail should be paid to the influence of adverse sample holding and storage conditions.¹⁴ For example, if a shipment of solid dosage tablets is stored in adverse conditions, there is a question if the product still can offer the same intended health benefit and if the product still is regarded as safe for consumption. The influence of heat and humidity are key factors in product stability.¹⁵ Given the sensitivity of the near infrared region to moisture, there is potential for detecting tablets stored under adverse conditions.

In recent years, technological advances have positioned handheld spectrometers in the market that are compact and come with a much lower cost-point than traditional benchtop spectrometers.¹⁶ However, these spectrometers are limited in the spectral range of data collection, the optical bandwidth, and typically come with the unknown influence of changes to the operating environment, such as increased temperature and humidity. Several detailed recent reviews can be found that describe the current benefits, conflicts, and unknowns concerning applied usage of these instruments in various fields such as agriculture, textiles, and pharmaceuticals.^17–19 Additionally, the use of qualitative analysis for falsified pharmaceutical tablets was undertaken with handheld Raman and near infrared (NIR) spectrometers.²⁰

In resource limited areas where it may be restrictive to screen essential medications for quality compliance with a benchtop DRS that requires a higher initial investment in funding, training, and software, it may be advantageous to implement handheld spectrometers for compliance screenings. Handheld and portable spectroscopy tools are not intended to replace traditional laboratory-based chemical testing and their confirmatory power, but are intended to compliment traditional methods when resources are limited.²¹ In this instance there is concern regards brand differentiation and determining whether environmentally stressed samples can be detected and flagged with DRS. A handheld spectrometer is used (in comparison to a benchtop spectrometer) to sample Levo-EE tablets containing two low-dose active ingredients, to determine if (a) brand discrimination of solid-dosage tablets is possible and (b) to determine if environmentally stressed tablets are rejected from a single-class screening method for one of the Levo-EE brands.

Materials and methods

Oral dosage tablets

Three brands of oral dosage tablets containing levonorgestrel and ethinyl estradiol were obtained from manufacturing sources for the initial brand qualitative discrimination. Oral contraceptives were obtained through routine screening from a donation based global health supply chain. Manufacturing sources are blinded for confidentiality purposes and are referred to as brands “A”, “B”, and “C” from here onwards. Products are approximately 75 mg per tablet, containing 30 µg of ethinyl estradiol and 150 µg of levonorgestrel. For the brand discrimination portion, 233 tablets of brand A were obtained from 8 lots, 120 tablets of brand B were obtained from 4 lots, additionally, 10 samples from one lot of brand A under a different label were collected, 108 tablets from 4 lots of brand C. Tablets come packaged with placebos that are typically taken after 21 days of taking the tablets containing the API, and are commonly either ferrous fumarate or lactose. Here, nine ferrous fumarate placebo tablets from brand A, B, C, and 3 lactose placebo tablets from brand A were also sampled and included as negative controls.

Unopened blister packs from a previous stability study containing Levo-EE tablets that were stored under adverse conditions were available for collecting spectra. This was done to compare Levo-EE tablets that had stored under elevated temperature or humidity conditions for an extended period of time. Tablets from the same manufacturing sources were placed into three environmental chambers (Thermo Scientific, model #3940, Marietta, OH, USA). One chamber was kept at 30 ± 2°C and approximately 75% relative humidity (RH), one at 40 ± 2°C and approximately 75% RH, and a third at 50 - 54 ± 2°C and 75% RH. Glass desiccators were also placed inside each environmental chamber with tablets stored in each to achieve a temperature effect but to keep tablets at a low level of humidity (labeled as 0% RH). All samples were kept in their original blister packaging and kept at either ambient conditions or in the environmental chambers. Table 1 details the number of samples and lots of each brand used as well as the tablets held under environmentally stressed conditions. Samples for brand A were stored in environmental chambers for 7 years and 1 month, while brand B samples were stored in environmental chambers for 5 years and 6 months. Including the brand discrimination portion with the samples stored under environmental stress, a total of 560 tablets were scanned with the benchtop and handheld spectrometers.

Table 1.

Levonorgestrel – ethinyl estradiol tablet samples scanned by a handheld spectrometer (900 – 1700 nm) and a benchtop spectrometer (350 – 2500 nm). Samples used in construction of the brand specific qualitative model for brand A were from tablets stored at ambient conditions.

Brand	Product	Storage condition	Lots #	Samples #
A	Levo-EE	Ambient	8	233
B	Levo-EE	Ambient	4	120
A	Levo-EE	Ambient	1	10
C	Levo-EE	Ambient	4	108
A/B/C	Ferrous fumarate	Ambient	3	9
C	Lactose	Ambient	1	3
A/B	Levo/EE	30°C/0% RH	2	14
A/B	Levo-EE	40°C/0% RH	2	14
A/B	Levo-EE	50 °C–54°C/0% RH	2	14
A/B	Levo-EE	30°C/75% RH	2	14
A/B	Levo-EE	40°C/75% RH	2	14
A/B	Levo-EE	50 °C–54°C/75% RH	1	7

Levo = levonorgestrel, EE = ethinyl estradiol, and RH = relative humidity.

Brand A tablets held under adverse environmental conditions were stored for 7 years and 1 month, Brand B tablets held under adverse environmental conditions were stored for 5 years and 6 months.

Spectrometers and data collection

The two NIR instruments used for spectral acquisition are shown in Figure 1 with the handheld spectrometer, a Tellspec Food Sensor (Tellspec Inc., Toronto, ON, Canada), shown in Figure 1(a) and the ASD LabSpec 5000 with Muglight^® assembly (Malvern Panalytical, Malvern, UK) shown in Figure 1(b). The portable handheld spectrometer operates in the short-wave infrared range (900 – 1700 nm) collecting data at a variable 2 – 4 nm for 256 points was used for scanning tablets. The Tellspec spectrometer consists of the NIR-S-G1 spectrometer (Innospectra Corp., Hsinchu, Taiwan), and operates through the free and downloadable version of the DC&M 2.0 software (Tellspec, Inc., Montreal, Canada), which was installed on to a mobile device for data collection, after being paired through Bluetooth^® with data stored in the app. For sample collection, a 12 × 35 mm borosilicate glass vial (FisherbrandTM 51, Fisher Scientific, Hampton, NH, USA) was cut in half using a motorized saw with a 19 mm or 38 mm rotary diamond blades (Dremel 4000, Racine, MI, USA). After cleaning the cut vial, it was affixed to the handheld spectrometer. Inside the vial, a 1.6 mm thick piece of Teflon sheeting (McMaster Carr, Elmhurst, IL, USA) was cut to the inside diameter of the vial and inserted. A #1 cork borer was used to cut a small hole (4 mm) in the center of the Teflon before inserting and was positioned directly over the scanning window of the handheld spectrometer. Background scans were collected with the empty vial and Teflon insert, and a second Teflon insert inserted into the vial to cover the small opening over the scanning window. For sample scans, the top piece of Teflon was removed, and tablets were positioned on top of the opening and scanning window. Tablets covered the opening in the Teflon completely. Triplicate scans were collected, and tablets were rotated approximately 45 degrees between scans. After scanning was complete, CSV files were exported as raw reflectance spectra. The background scans were used to adjust the raw reflectance and convert to absorbance Log₁₀(1/R) for further analysis, where R = I/I_o with I for the sample data and I_o for the reference. NIR spectra of standards for levonorgestrel, ethinyl estradiol, and lactose (purity 97.6% or higher) (Sigma Aldrich, St Louis, MO, USA) were acquired by placing the powders in individual 12 mm × 35 mm borosilicate glass vials (Fisher Scientific, Waltham, MA, USA). A background was taken by another of the previously mentioned vials and cutting a piece of 1.6 mm thick Teflon to the inside diameter and placing in the bottom of the vial, then scanning in triplicate against the sample window.

Figure 1.

(a) Tablet scanning adapter and setup for use with the handheld spectrometer (900 – 1700 nm). (b) Tablet scanning setup for the benchtop spectrometer using the Muglight® attachment, sample tray adapter, and Spectralon cover for tablet data collection.

The Labspec 5000 was used for scanning the tablets from 350 to 2500 nm collecting data every 1 nm for 2150 points. The Muglight^® accessory was used to collect data with a small sample tray adapter. First, a background was collected as the baseline using a 25.4 mm Spectralon puck inserted into the sample tray. Tablets were centered on the sample tray adapter and covered with a 89 mm Spectralon disk (Malvern Panalytical, Malvern, UK). Scans were collected in triplicate with tablets rotated approximately 45 degrees between scans. Each spectra was an average of 50 scans, resulting in a final spectra averaging 150 scans. Spectra were saved as Log₁₀(1/R) and exported as an ASCII file. The three secondary reference standards for levonorgestrel, ethinyl estradiol, and lactose were scanned in a similar fashion as described with the handheld, but with the previously mentioned Muglight^® attachment for the benchtop instrument. Here, another piece of 1.6 mm thick Teflon was cut to the inside diameter of the sample tray adapter with a hole cut in the center, the same diameter as the sample vial. References vials (previously described for the handheld) were also scanned in this manner for the benchtop spectrometer.

Multivariate data analysis

After converting raw reflectance into Log₁₀(1/R), data were imported into R-studio (2022.07.01) (Boston, MA, USA), where all analyses were conducted. A mean-centered singular value decomposition (SVD) based principal component analysis (PCA) with a leave-one-out cross validation was conducted on samples from both the benchtop spectrometer dataset and separately with the handheld spectrometer’s dataset using an R-Studio Shiny App.²² A single product specific multivariate classification model was developed for “brand A”. A classification model built around Mahalanobis distances (M-dist.) based on principal component (PC) scores (mean-centered) and residuals was established for brand A as the reference dataset. In brief, reference samples were used to build a spectral library, and samples of other brands, placebos, and tablets stored under environmental stress were projected against the model to determine if the benchtop and portable spectrometers could differentiate negative controls from the reference, and verify the model with positive controls of new tablets from brand A. Reference and positive/negative controls were split amongst lots to approximately 70 % for references and 30 % for positive controls. More information on the M-dist. classification approach and establishing the thresholding value for classification at a 95% confidence interval can be found in Gemperline and Boyer and Kaale et al.^23,24

Results and discussion

Brand comparison

The mean spectra by brand for tablets stored at ambient conditions for the benchtop is seen in Figure 2(a) and handheld spectrometer is shown in Figure 2(b). The spectra for brands A, B, and C are similar in pattern and intensity, whereas the ferrous fumarate and lactose placebos are noticeably different with spectra obtained from the benchtop spectrometer. Spectral differences are more pronounced with the handheld spectrometer’s mean spectra showing that brands B and C are different around the 1000 nm region from brand A. The ferrous fumarate and lactose placebos differ from the spectra of brands A, B, and C by a substantial amount, and are easily differentiated, visibly. The spectra for reference standards are shown for the benchtop spectrometer in Figure 2(c) and the handheld spectrometer in Figure 2(d). The variable 2 – 4 nm bandwidth of the handheld spectrometer is noticeably different from the benchtop’s 1 nm bandwidth, as the spectral peaks are broader for the handheld than the benchtop. Each of the three reference standards have definitive spectral peaks that are also noticed in the benchtop spectra. Lactose peaks at 1543 nm and 1593 nm are noticeable, while levonorgestrel peaks found at 1185 nm, 1386 nm, and 1558 nm. While ethinyl estradiol peaks can be found at 1420 nm, 1468 nm, and 1543 nm are observed with both spectrometers. However, there are multiple key spectral features for each of the three standards found beyond the 1700 nm limit of the handheld spectrometer that are captured by the benchtop spectrometer.

Figure 2.

Mean spectra of tablets stored at ambient conditions by brand/placebo for both the benchtop spectrometer (a) and the handheld spectrometer (b). Reference spectra for lactose (a common excipient), levenorgesterol (levo), and ethinyl-estradiol (EE) are shown for the benchtop spectrometer (c), and the handheld spectrometer (d).

Principal component analysis was applied to the dataset containing the three brands of Levo-EE stored at ambient conditions. The PCA scores plot can be seen in Figure 3 with the benchtop spectrometer shown in the top row, Figure 3(a) and (b), and handheld spectrometer shown in the bottom row, Figure 3(c) and (d). The PCA scores plots represent similarities and dissimilarities between samples of Levo-EE. Brand A was established as the reference samples with positive controls also from brand A. Negative controls 1 (NC1) (brand B) and 2 (NC2) (brand C) are two differently labeled brands but are understood to be produced by the same manufacturer and packaged with two different labels. Considering this, it explains why the NC1 and NC2 samples overlap each other on the PCA scores plots for both spectrometers. Also, the PC samples overlap the reference (Ref) samples showing consistency and potential for single class model sensitivity. Score values from the benchtop spectrometer show an approximately normal distribution of all samples and cluster overlapping when comparing PC1 (72.23%) and PC2 (16.75%). Plotting lower PCs shows a clear separation between Ref/PC and the NC1/NC2 samples, which can be seen by the plot in Figure 3(b). The handheld spectrometer resulted in PCA score values being separated with PC1 (71.5%) and PC2 (24.2%). Whereas the benchtop spectrometer relates more influence from the visible light range, with tablet color likely having a greater impact on PC1.

Figure 3.

Scores plots from principal component analyses of data collected with the benchtop spectrometer (a and b) and the handheld spectrometer (c and d) for brands A, B, and C Benchtop spectrometer results shown in the top row and handheld spectrometer shown in the bottom row. Ref = reference, PC = positive control, NC1/NC2 = negative controls 1 and 2.

Comparing PC loading vectors gives an indication to the variable-to-variable implications on the PCA scores. The benchtop spectrometer’s PCA loading vectors for PCs 1 – 4 (explaining 98.7% of the PCA model variance) are shown in Figure 4(a) with PC1 shown as an insert, due to scaler differences in the y-axes. Primarily, PC1 is similar in pattern to that of the Levo-EE mean spectra. From the scores plot, we can identify PC3 as being a key contributing PC to the brand discernability between Levo-EE tablets. A loading peak at 1143 nm is noticeably influencing PC3, and is noted in the previously shown reference standards’ spectra for levonorgestrel and ethinyl estradiol. Another loading peak at 1386 nm is impacting PC3 and potentially correlates to spectral peaks noticed in levonorgestrel, ethinyl-estradiol, and lactose reference spectra. The loading vectors from the handheld spectrometer are shown in Figure 4(b). PCs 1- 3 are shown here and explain a total of 98.74% of the handheld model’s variance. PC2 was identified as a critical PC in qualitative differentiation between brands of Levo-EE. Within PC2 there is an influential loading peak at 1386 nm, and correlating to a spectral peak found in the levonorgestrel, ethinyl-estradiol, and lactose reference spectra. This loadings peak was also noticed within the benchtop loadings. Further approximate assignments based on collected reference scans for Levo, EE, and lactose are listed in Table 2.

Figure 4.

Loading vectors for principal components 1 – 4 for principal component analysis of benchtop spectrometer with the PC1 loading vector inserted (a) and the handheld spectrometer (b) for PCs 1 - 3. Samples included in PCA are for the brand discrimination section consisting of in-date and ambient stored brands A, B, and C only.

Table 2.

Proposed spectral peaks from levonorgestrel (Levo), ethinyl-estradiol (EE), and lactose secondary reference standards collected across the Labspec range (350 – 2500 nm).

Peak (nm)	Standard
1143	Levo, EE
1185	Levo, EE
1212	Lactose
1386	Levo, EE, lactose
1420	EE
1468	EE
1536	Lactose
1543	EE
1593	Lactose
1688	EE
1696	Lactose
1716	Levo, EE
1730	EE, lactose
1765	Lactose
1811	Levo
1834	EE, lactose
1936	Lactose (not desiccated)
1981	EE
2001	Levo
2093	Lactose, EE, levo
2134	Levo, EE
2170	EE
2289	Levo, EE
2335	Levo, EE, lactose
2386	Levo, EE

Environmental stress and brand classification

Once tablets were removed from the environmental chambers, they were visually inspected while still inside of the blister packs. Samples for both brands stored at 30°C and 40°C at 0% RH were visually similar, while brand B stored at 50 °C–54°C and 75% RH appeared dark red, and once the blister packs opened, tablets turned to a paste upon removing tablets from the package, having lost physical stability. These tablets were unable to be scanned with either spectrometer. Brand A stored under the same conditions had a slight yellowish hue, but tablets maintained their physical integrity and were able to be scanned by benchtop and handheld spectrometers. The mean spectra from tablets that were stored under adverse environmental conditions can be seen in Figure 5 (under desiccation) and Figure 6 (75 % RH). Spectra shown here are not preprocessed and we can see that there is a scaler difference for brand A and B with samples stored at 50 °C–54°C resulting in a lower Absorbance Log₁₀(1/R) spectra for both the benchtop and handheld spectra. Samples stored at 30°C and 40°C showed more similarities between mean spectra. The mean spectra for brand B at 30 and 40°C are shown, but spectra for 50 °C–54°C are missing from brand B, because the tablets absorbed enough moisture to deteriorate the tablets in a way that rendered them unusable. Brand B shows that there is a greater change in spectra at and above the peak shown for moisture peak at approximately 1940 nm, however, this change is less pronounced for brand A. Brand A samples stored at 75% RH also shows a greater difference between tablets stored at 50 °C–54°C than 30 or 40°C, similar to the temperature correlations for desiccated samples. Samples stored at 50 °C–54°C also resulted in lower spectral peak resolutions with broader peaks noticed. This would also suggest that tablets stored in the range of 30 °C–40°C do exhibit noticeable changes in their spectra, but tablets stored at 50 °C–54°C have a noticeably larger change in spectral pattern. With these tablets stored in environmental chambers, we are observing spectral changes under long-term storage with adverse heat and humidity conditions, representing a possible worst-case scenario.

Figure 5.

Mean spectra for benchtop spectrometer (a and b) and handheld spectrometer (c and d) for tablets stored under desiccation at 30, 40, or 50°C–54°C. Brand A shown in a and c, while brand B is shown in (b and d).

Figure 6.

Mean spectra for benchtop spectrometer (a and b) and handheld spectrometer (c and d) for tablets stored at 75 % relative humidity and at temperatures of 30, 40, or 50 °C–54°C. Brand A shown in a and c, while brand B is shown in (b and d).

Moisture influence

Solid dosage tablets can have a shelf life of several years, and therefore preventing moisture pickup during storage is a primary concern during manufacturing and packaging. Previously, our group obtained blister packs of swollen and cracked oral contraceptive tablets provided through USAID’s program. Tablets were scanned with the benchtop spectrometer and results are shown in Figure S1. These results were previously unpublished but offer insight into the moisture retention in the tablets and changes to the spectra. Figure S1(a) shows that the visibly noticeable cracked tablets inside the blister packs, while Figure S1(b) shows the number of tablets cracked or damaged for each of the four lots, where the vast majority of cracked tablets occurred within cells that showed no evidence of indentation. The mean raw spectra of damaged and non-damaged tablets can be seen in Figure S1(c). The water overtone peak at approximately 1940 nm, which is emphasized by performing a 1st derivative Savitzky-Golay 1st preprocessing step shown in Figure S1(d).

After removing the tablets for brands, A and B from storage in the environmental chambers and scanning, spectra were also derivatized (1st derivative Savitzky-Golay of Absorbance Log₁₀(1/R)) and plotted to observe the spectral changes in each brand by changes in storage conditions, with brand A shown in Figure 7 and brand B shown in Figure 8. For each figure, the benchtop spectrometer’s resulting spectra is shown on the left (a) (1000 – 2500 nm), while the handheld spectrometer’s results are shown on the right (b) (900 – 1700 nm). Similar to previous work (Figure S1), derivatized spectra show evidence of moisture at different temperatures/humidity conditions (relative to ambient conditions) in the regions of 1930-1940 nm for both brands and spectrometers. Recently, Patel, et al., (2023) showed that moisture in tablets could be quantified using a handheld near-infrared spectrometer (1596 – 2396 nm).²⁵

Figure 7.

Mean spectra of samples stored under 75 % relative humidity for brand A tablets. Boxes highlight the water absorbance ranges. (a) The benchtop spectra with a 1st derivative Savitzky-Golay transform (1st order polynomial, and a window size of 5). (b) The handheld spectra with a 1st derivative Savitzky-Golay transform (1st order polynomial, and a window size of 3).

Figure 8.

Mean spectra of samples stored under 75 % relative humidity for brand B tablets. Boxes highlight the water absorbance ranges. (a) The benchtop spectra with a 1st derivative Savitzky-Golay transform (1st order polynomial, and a window size of 5). (b) The handheld spectra with 1st derivative Savitzky-Golay transform (1st order polynomial, and a window size of 3).

Multivariate classification results based on an M-dist. method for brand A can be seen in Table 3. The PCA model was reduced to 12 PCs, with an M-dist. thresholding value for the brand A model established at 4.84 from a 95% confidence interval. The handheld spectrometer PCA model established an M-dist. threshold at 3.66 and 6 PCs at a 95% confidence interval.

Table 3.

Mahalanobis distance results for a qualitative discriminatory method for product A using the benchtop and handheld spectrometers.

Samples	Benchtop	Benchtop	Handheld	Handheld
Samples	Mean ± StDev	Correct (%)	Mean ± StDev	Correct (%)
	PCs: 12, Threshold: 4.84		PCs: 6, Threshold: 3.66
Reference (A)	0.97 ± 0.22	160/160 (100%)	0.91 ± 0.41	159/160 (99.38%)
PC (A)	2.86 ± 0.68	73/73 (100%)	2.38 ± 0.39	73/73 (100%)
NC1 (B)	283.7 ± 105.5	120/120 (100%)	9.84 ± 0.3	120/120 (100%)
NC2 (C)	249.2 ± 84.2	108/108 (100%)	9.51 ± 0.27	108/108 (100%)
FF (A,B,C)	1921.66 ± 632.74	9/9 (100%)	12.62 ± 0.68	9/9 (100%)
Intermediate (A)	3.05 ± 0.71	10/10 (100%)	8.34 ± 0.06	0/10 (0%)
Lactose (A)	113.32 ± 7.14	3/3 (100%)	8.71 ± 0.04	3/3 (100%)
0RH_30C (A)	4.65 ± 0.88	2/7 (28.57%)	9.92 ± 0.67	7/7 (100%)
0RH_30C (B)	277.77 ± 25.96	7/7 (100%)	14.73 ± 0.74	7/7 (100%)
0RH_40C (A)	6.85 ± 0.75	7/7 (100%)	9.55 ± 0.38	7/7 (100%)
0RH_40C (B)	308.07 ± 65.45	7/7 (100%)	14.93 ± 1.5	7/7 (100%)
0RH_50C (A)	10.71 ± 1.11	7/7 (100%)	7.85 ± 0.08	7/7 (100%)
0RH_50C (B)	227.3 ± 61.93	7/7 (100%)	10.04 ± 0.5	7/7 (100%)
75RH_30C (A)	6.97 ± 1.39	7/7 (100%)	9.96 ± 0.93	7/7 (100%)
75RH_30C (B)	316.69 ± 94.19	7/7 (100%)	17.19 ± 0.84	7/7 (100%)
75RH_40C (A)	48.29 ± 15.74	7/7 (100%)	9.98 ± 0.37	7/7 (100%)
75RH_40C (B)	276.28 ± 99.11	7/7 (100%)	12.73 ± 1.36	7/7 (100%)
75RH_50C (A)	698.42 ± 395.84	7/7 (100%)	65.83 ± 16.52	7/7 (100%)

A or B refer to the brand of levonorgestrel and ethinyl estradiol.

All the benchtop reference samples, and 159 of the 160 handheld samples resulted in a passing classification for the brand A specific method, while all of the PCs (n = 73) tested as passing for both spectrometers. The one misclassified handheld tablet can be seen as an outlier in the PCA score plot in Figure 3(c). All brand specific negative controls (NCs) stored at ambient conditions as well as the environmentally stressed tablets listed as NCs failed the method for brand A with M-dist. Values above the respective thresholds for the two spectrometers. There was one exception for brand A tablets stored under desiccation at 30°C, which saw 5 of the 7 tablets pass for the brand A stored at ambient temperatures, though be it by a close mean of 0.19. Interestingly, the same seven tablets failed the handheld method as expected. This is the least drastic storage condition, imparting no real color change on the tablets. It may be possible that these five tablets barely passed the classification method on the benchtop spectrometer because it also collects wavelengths in the VIS range, which may be potentially not causing a difference in the spectra, but factor into the PCA and resulting M-dist. calculation, versus the handheld which only collects data in the short-wave infrared range. There is benefit in collecting the VIS range of spectra by the benchtop, and why it was included in this analysis. Here, we want an understanding of the capabilities and limitations of each spectrometer, not just a comparison of the same spectral range. There is one lot (10 tablets) listed as an intermediate control, and were collected from brand A, but were packaged with a different placebo (lactose not ferrous fumarate). The benchtop spectrometer results passed the M-dist. method as brand A, though be it at a higher level (mean = 3.05), but the handheld spectrometer resulted in all 10 tablets failing. There are production questions concerning this lot, which is why it was listed as an intermediate control. Because the benchtop spectrometer is collecting VIS wavelengths that are equally weighted in the PCA and resulting M-dist., this may also factor into why the benchtop is passing, and the handheld is not resulting in a passing value. Overall, both spectrometers resulted in promising accuracy with the benchtop spectrometer correctly classifying 555 out of 560 tablets = 99.12 % accuracy, while the handheld spectrometer classified 549 of 560 tablets = 98.04 % accuracy. Here, the challenge with Levo-EE tablets is that there are two low-dose active ingredients present in relatively small tablets. Considering that the handheld spectrometer resulted in accuracy >98 % for differentiating between brands as well as tablets stored under environmental stressors, there is promise in adapting similar quality screening methods for other solid dosage tablets using low-cost handheld methods, potentially opening up screening opportunities in resource limited areas (as previously demonstrated with depot-medroxyprogesterone acetate).²⁶ While the benchtop does collect spectra in a greater range, capturing possible color changes in the VIS range, which can be indictive of stability issues, and additional Levo-EE related peaks up to 2500 nm, it serves as an efficient spectroscopy screening method for Levo-EE, however, it is limited in cost and portability. The handheld DRS does not collect as much spectral information but has shown the ability to differentiate between brands.

Conclusions

In low- and middle-income countries, sub-standard and counterfeit pharmaceuticals are an ongoing concern. Portable and low-cost screening methods for essential medicines have the potential for impactful quality compliance systems. Here, a handheld spectrometer collecting data in the short-wave infra-red spectral range was able to qualitatively differentiate between brands of Levo-EE. When additional Levo-EE samples stored under environmental stress for an extended period of time were compared against the model for brand “A”, the stressed tablets were easily identified as different from the reference model for tablets stored under ambient conditions. The spectra collected from the environmentally stressed tablets had noticeable increases at known near infrared water absorbance peaks at approximately 1450 nm and 1940 nm. Not only were spectral changes noted on the benchtop spectrometer, but was also picked up by the handheld spectrometer, suggesting that the handheld spectrometer is not only sensitive enough to detect brand differences in tablets containing two low-dose active ingredients, but also sensitive enough to detect tablets that have been under environmental stress, potentially reducing the efficaciousness of the medicine.

Supplemental Material

Supplemental Material - Qualitative discrimination of low-dose oral contraceptives and identification of environmentally stressed tablets: Comparing a handheld near-infrared spectrometer (900 – 1700 nm) to a benchtop spectrometer (350 – 2500 nm)

Supplemental Material for Qualitative discrimination of low-dose oral contraceptives and identification of environmentally stressed tablets: Comparing a handheld near-infrared spectrometer (900 – 1700 nm) to a benchtop spectrometer (350 – 2500 nm) by Matthew Eady, Noah Peters and David Jenkins in Journal of Near Infrared Spectroscopy.

Footnotes

Acknowledgements

This work was supported through Contract No. AID-OAA-C-15-00001, USAID, Global health Supply Chain Quality Assurance (implemented by FHI 360). Also, FHI 360 (Product Quality and Compliance Department) Research Associate Daniel Highland for assistance with preparing figures.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: United States Agency for International Development; AID-OAA-C-15-00001.

ORCID iDs

Matthew Eady

David Jenkins

Supplemental Material

Supplemental material for this article is available online.

References

World Health Organization (WHO) . Model list of essential medicines 2021. https://list.essentialmeds.org/. Accessed 16 November 2021.

Vivalo

Bornbusch

US Agency for International Development (USAID). Overview of contraceptive and condom shipments. Washington D.C., USA: US Agency for International Development (USAID) - Office of Population Health, 2020. https://www.usaid.gov/sites/default/files/documents/CC_FY20.pdf. Accessed, December 6, 2021.

United Nations, Department of Economic and Social Affairs, Population Division . Contraceptive use by method 2019 data booklet 2019.

Torgrimson

Meendering

Kaplan

, et al. Endothelial function across an oral contraceptive cycle in women using levonorgestrel and ethinyl estradiol. Am J Physiol Heart Circ Physiol 2007; 292: H2874–H2880. DOI: 10.1152/ajpheart.00762.2006.

Newton

Amin

Bird

, et al. The primacy of public health considerations in defining poor quality medicines. PLoS Med 2011; 8: e1001139. DOI: 10.1371/journal.pmed.1001139.

Monge

Dwivedi

Zhou

, et al. A tiered analytical approach for investigating poor quality emergency contraceptives. PLoS One 2014; 9(4): e95353. DOI: 10.1371/journal.pone.0095353.

Ozawa

Evans

Bessias

, et al. Prevalence and estimated economic burden of substandard and falsified medicines in low- and middle-income countries: a systematic review and meta-analysis. JAMA Netw Open 2018; 1(4): e181662. DOI: 10.1001/jamanetworkopen.2018.1662.

Nayyar

Breman

Herrington

. The global pandemic of falsified medicines: laboratory and field innovations and policy perspectives. Am J Trop Med Hyg 2015; 92: 2–7. DOI: 10.4269/ajtmh.15-0221.

Igne

Ciurczak

. Near-infrared spectroscopy in the pharmaceutical industry. In: Ozaki

Huck

Tsuchikawa

, et al. (eds) Near-infrared spectroscopy: theory, spectral analysis, instrumentation, and applications. Singapore: Springer, 2021a, pp. 391–412.

10.

Pino-Torres

Maspoch

Perez-Rivra

, et al. Evaluation of NIR and Raman spectroscopies for the quality analytical control of a solid pharmaceutical formulation with three active ingredients. Microchem J 2020; 154: 104576. DOI: 10.1016/j.microc.2019.104576.

11.

Sacre

P-Y

Bleye

Hubert , et al. PAT Applications of NIR spectroscopy in the pharmaceutical industry. In: Crocombe

Leary

Kammrath

(eds) Portable spectroscopy and spectrometry 1: technologies and instrumentation. Hoboken, NJ: Wiley, 2021, pp. 67–88. DOI: 10.1002/9781119636489.ch4.

12.

Igne

Ciurczak

. Pharmaceutical analysis. In: Ciurczak

Igne

Workman

J, J

, et al. (eds) Handbook of Near-Infrared Analysis. Boca Raton, FL, USA: CRC Press, 2021, pp. 817–837.

13.

Blanco

Alcala

Gonzalez

, et al. A process analytical technology approach based on near infrared spectroscopy: tablet hardness, content uniformity, and dissolution test measurements of intact tablets. J Pharmaceut Sci 2006; 95(10): 2137–2144. DOI: 10.1002/jps.20653.

14.

Gbenga

Taiwo

. Studies of the effect of storage conditions on some pharmaceutical parameters of powders and tablets. Dhaka Univ J Pharm Sci 2015; 14(2): 147–151.

15.

Buehler

Huynh-Ba

. Regulatory perspectives on product stability. In: Huynh-Ba

(ed) Pharmaceutical stability testing to support global markets. Berlin/Heidelberg, Germany: Springer Science and Business Media, 2010. DOI: 10.1007/978-1-44199-0889-6_2.

16.

Crocombe

. Portable spectroscopy. Appl Spectrosc 2018; 72(12): 1701–1751. DOI: 10.1177/0003702818809719.

17.

Yan

Siesler

. Quantitative analysis of a pharmaceutical formulation: performance comparison of different handheld near-infrared spectrometers. J Pharm Biomed Anal 2018; 160: 179–186. DOI: 10.1016/j.jpba.2018.07.048.

18.

Alcala

Blanco

Moyano

, et al. Qualitative and quantitative pharmaceutical analysis with a novel hand-held miniature near infrared spectrometer. J Near Infrar Spectrosc 2013; 21: 445–457. DOI: 10.1255/jnirs.1084.

19.

Deidda

Sacre

Clavaud

, et al. Vibrational spectroscopy in analysis of pharmaceuticals: critical review of innovative portable and handheld NIR and Raman spectrophotometers. Trends Anal Chem 2019; 114: 251–259. DOI: 10.1016/j.trac.2019.02.035.

20.

Ciza

Sacre

Waffo

, et al. Comparing the qualitative performances of handheld NIR and Raman spectrophotometers for the detection of falsified pharmaceutical products. Talanta 2019; 202: 469–478. DOI: 10.1016/j.talanta.2019.04.049.

21.

U.S. Pharmacopeia - National Formulary . Evaluating screening technologies for assessing medicine quality 1850.

22.

Sortijas

. Github. PCA-MDR 2021. https://github.com/sortijas/MDA-PCA.

23.

Gemperline

Boyer

. Classification of near-infrared spectra using wavelength distances: comparison to the Mahalanobis distance and residual variance methods. Anal Chem 1995; 67: 160–166.

24.

Kaale

Hope

Jenkins

, et al. Implementation of 350-2500 nm diffuse reflectance spectroscopy and high-performance thin-layer chromatography to rapidly assess manufacturing consistency and quality of cotrimoxazole tablets in Tanzania. Trop Med Int Health 2015; 21: 61–69. DOI: 10.1111/tmi.12621.

25.

Patel

Jin

Handzo

, et al. Measurement of moisture content in pharmaceutical tablets by handheld near-infrared spectrometer: adopting quality by design approach to analytical method lifecycle management. J Pharm Biomed Anal 2023; 229: 115381.

26.

Eady

Payne

Sortijas

, et al. A low-cost and portable near-infrared spectrometer using open-source multivariate data analysis software for rapid discriminatory quality assessment of medroxyprogesterone acetate injectables. Spectrochim Acta 2021; 259: 119917. DOI: 10.1016/j.saa.2021.119917.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.08 MB