A framework for selecting analytical techniques in profiling authentic and counterfeit Viagra and Cialis

https://doi.org/10.1016/j.forsciint.2013.12.005Get rights and content

Abstract

Several analytical techniques aimed at profiling drugs are deemed costly and time consuming, and may not be promptly available for analysis when required. This paper proposes a method for identifying the analytical techniques providing the most relevant data for classification of drug samples into authentic and unauthentic categories. For that matter, we integrate principal components analysis (PCA) to k-Nearest Neighbor (KNN) and Support Vector Machine (SVM) classification tools. PCA is first applied to data from five techniques, i.e., physical profile, X-ray fluorescence (XRF), direct infusion electrospray ionization mass spectrometry (ESI-MS), active pharmacological ingredients profile (ultra performance liquid chromatography, UPLC–MS), and infrared spectroscopic profile (ATR-FTIR). Subsets of PCA scores are then combined with a “leave one subset out at a time” approach, and the classification accuracy using KNN and SVM evaluated after each subset is omitted. Subsets yielding the maximum accuracy indicate the techniques to be prioritized in profiling applications. When applied to data from Viagra and Cialis, the proposed method recommended using the data from UPLC–MS, physical profile and ATR-FTIR techniques, which increased the categorization accuracy. In addition, the SVM classification tool is suggested as more accurate when compared to the KNN.

Introduction

The production of counterfeit medicines is a criminal problem that carries serious risks to public health [1], since there is no certainty about pharmaceutical dosage forms, active pharmacological ingredient, and origin of raw materials or manufacturing conditions. Such problem has motivated the development and application of a large number of analytical techniques and multivariate tools in forensic analysis tailored at discriminating between authentic and unauthentic samples of seized drugs, and finding similar properties in unauthentic samples [2], [3].

In recent years, our group has applied profiling approaches and image processing in counterfeit Viagra® (sildenafil citrate, SLD, Pfizer) and Cialis® (tadalafil, TAD, Eli Lilly) samples seized by the Brazilian Federal Police in the state of Rio Grande do Sul, Southern Brazil. Such studies have relied on five analytical techniques, which were later integrated to multivariate techniques. The first analytical technique is the assessment of tablets physical profile [4] as suggested in [5], [6], which provided 4 post-tabletting variables: mass (mg), thickness (mm), shorter length (mm) and longer length (mm). Such variables were analyzed through the F-test (ANOVA), showing that counterfeiters cannot mimic the mass variable in unauthentic drugs, but can often mimic genuine products length. The second analytical technique provided inorganic fingerprinting data obtained by X-ray fluorescence (XRF), originating 2048 variables (energy, keV) [7]. XRF is a nondestructive technique for characterization of metal presence [8] featured by multielemental capability, good detectivity, and short analysis time. In our propositions, XRF was aligned with PCA and hierarchical cluster analysis to enable the semi-quantitative determination of sildenafil citrate and excipients such as calcium phosphate, titanium oxide and iron oxide. That allowed us to classify authentic and counterfeit Cialis and Viagra samples.

The third analytical technique, organic profile achieved by direct infusion electrospray ionization mass spectrometry (ESI-MS), highlights the polar composition of seized tablets [9] without the need for isolation and prior chromatographic separation. Such analysis generated 99 variables corresponding to ions (m/z) in methanol extracts samples. Spectra for authentic Viagra® depicted ions exclusively corresponding to the SLD molecule: [SLD + H]+ of m/z 475; [SLD + Na] + of m/z 497; and [2SLD + H]+ of m/z 949; authentic Cialis® showed ions of m/z 343, 365 and 707 from the lactose molecule (excipient). PCA was applied to ESI-MS fingerprint data, allowing the classification of samples into categories based on their contents of active ingredients. The fourth analytical technique, UPLC–MS [10], enabled qualitative and quantitative determination of the active pharmaceutical ingredients (APIs), and yielded 3 variables: SLD content, HSD (Homosildenafil) content and TAD content (in mcg/mg). Such variables enabled assessing products authenticity, presence of the API indicated on the label, API concentration and presence of other contaminants. Finally, the fifth analytical technique relied on attenuated total reflection Fourier transform infrared (ATR-FTIR) [11]. This experiment generated 661 variables (wavenumbers, cm−1) on mid-infrared region (1800–525 cm−1), which includes the absorption region that highlights major differences in PDE-5 inhibitors [12]. PCA was applied to ATR-FTIR aimed at grouping samples according to chemical profiles, distinguishing successfully between authentic and counterfeits samples and suggesting a common illicit source for different seized batches. Although not assessed in this paper, other relevant analytical techniques and multivariate tools have been employed to detect counterfeit drug samples. Degardin et al. [13] applied Support Vector Machine and PCA to classify and profile data from Raman spectra, while Deconinck et al. [14] performed an exploratory analysis based on projection pursuit and clustering aimed at discriminate between authentic and counterfeit samples of Viagra and Cialis. Similarly, Been et al. [15] applied several supervised and unsupervised techniques to near-infrared (NIR) and Raman spectroscopy data, including KNN, partial least squares discriminant analysis and probabilistic neural network, while Deconinck et al. [16] suggested a framework based on classification trees to categorize authentic and unauthentic medicine samples. Finally, Anzanello et al. [3] proposed a wavenumber selection approach to identify the most relevant FTIR bands for inserting Viagra and Cialis samples into two classes, and Sacré et al. [17] used partial least squares to identify the most effective FITR bands for detecting forged medicines.

Among the several benefits provided by the five aforementioned analytical techniques in forensic applications, two are noteworthy: (i) they enable detecting authentic and unauthentic samples of seized medicines; and (ii) they allow finding similar properties in unauthentic samples, making it possible for police forces to unveil identical sources from different drug seizures. The availability and cost of such techniques, however, may limit the ability to run the entire set of analyses, since many laboratories and forensic institutes do not have the necessary equipment to perform all techniques for a consolidated result. In addition, reduced time for obtaining conclusive results is typically verified in practical cases, prohibiting the running of all experiments. Thus, it seems reasonable developing methods able to identify the analytical techniques responsible for providing the most relevant data aimed at inserting seized samples into proper categories.

This paper proposes a method for selecting the analytical techniques providing the most conclusive data for categorizing seized drugs into authentic and unauthentic classes. For that matter, we integrate PCA to two data mining tools with classification purposes, k-Nearest Neighbor (KNN) and Support Vector Machine (SVM). In our propositions, PCA is applied to the data provided by the five analytical techniques, and the subsets of PCA scores deriving from each analytical technique are compiled into a single matrix consisting of new classificatory variables. The use of PCA is justified by its ability to merge information from analytical techniques described by a different number of variables into a new set of variables (the scores) similar in number and magnitude. Next, the most efficient subsets of PCA scores (each subset related to an analytical technique) are identified combining a “leave one subset out at a time” procedure with the KNN and SVM. In such procedure, each subset of scores is momentarily omitted from the dataset, a categorization using the remaining scores subsets is performed, and the classification accuracy, i.e., the proportion of correct categorizations, is evaluated. When all subsets were omitted once, the subset yielding the highest accuracy is removed from the dataset since it is the one that contributes the least in separating samples into categories. This iterative procedure is performed in the remaining subsets until there is only one subset left. The maximum accuracy obtained during the elimination process indicates the analytical techniques that provide the most relevant data for accurate classifications. Such techniques are then recommended in time and budget limited scenarios, and may be used as a guide to acquire new equipments for future analyses.

When applied to Viagra and Cialis data derived from the five aforementioned techniques, the proposed method recommended the use of data provided by UPLC–MS, physical profile and ATR FTIR techniques, since such data increased the classification accuracy of samples into authentic and counterfeit classes. Further, the SVM classification tool was suggested as a more accurate technique when compared to the KNN algorithm.

Section snippets

Samples

Fifteen counterfeit Viagra tablets from four different seizures, 36 counterfeit Cialis tablets from five different seizures, 4 authentic tablets of authentic Viagra®, and 4 tablets of authentic Cialis® were sent to the Technical and Scientific Division for Forensic Analysis of Rio Grande do Sul State. Sildenafil citrate (99.9%) and Viagra® tablets containing 50 mg of SLD were supplied by Pfizer Ltda Laboratories. Tadalafil (99.8%) and Cialis® tablets containing 20 mg of TAD were supplied by Eli

Results and discussion

The proposed method was applied to data provided by five analytical techniques used to profile seized Viagra and Cialis. All computational procedures were performed on a Matlab 7.8. The analytical techniques are coded as follows: (A) physical profile, (B) inorganic profile – XRF analysis, (C) organic profile – direct infusion ESI-MS analysis, (D) active pharmacological ingredients profile – UPLC–MS analysis, and (E) infrared spectroscopic profile – ATR FTIR analysis. That code is from now on

Conclusions

The increasing commerce and use of counterfeit medicines offers serious risks to public health worldwide, and the repression to that commerce mobilizes health surveillance agents, police and forensic forces. Several of the analytical techniques used for profiling seized drugs tend to be costly and restrict in terms of availability.

This paper proposed a framework for selecting the analytical techniques providing the most conclusive data for categorizing seized drugs into authentic and

References (31)

Cited by (18)

  • Combining wavelength importance ranking to the random forest classifier to analyze multiclass spectral data

    2021, Forensic Science International
    Citation Excerpt :

    The need for product traceability is also of utmost interest, as it may reflect on product reputation and market value. As for pharmaceutical products, tampered medicines represent serious risks to public health due to the lack of information about pharmaceutical dosage forms, origin of raw materials and active pharmacological ingredients, and manufacturing conditions [9,10]. That problem is specially observed in phosphodiesterase type 5 (PDE-5) inhibitors given the embarrassment associated with the pathology, and the easy purchasing of tampered products from fraudulent websites [11].

  • Ion beam analysis (IBA) and instrumental neutron activation analysis (INAA) for forensic characterisation of authentic Viagra® and of sildenafil-based illegal products

    2021, Talanta
    Citation Excerpt :

    It was the first oral pharmaceutical product to treat men with ED by inhibiting the type-5 phosphodiesterase (PDE5) enzyme [7]. Forensic characterisation of counterfeit Viagra® can be carried out by both image processing [8] and analytical techniques [9]. The analytical approaches tested include liquid chromatography-mass spectrometry (LC-MS) [10–15], nuclear magnetic resonance (NMR) [16,17], Fourier transform infrared spectroscopy (FTIR) [18] and Raman micro-spectroscopy [19].

  • Handbook of Analytical Techniques for Forensic Samples: Current and Emerging Developments

    2020, Handbook of Analytical Techniques for Forensic Samples: Current and Emerging Developments
  • Instrumental neutron activation analysis (INAA) and liquid chromatography (LC) coupled to high resolution mass spectrometry (HRMS) characterisation of sildenafil based products seized on the Italian illegal market

    2019, Forensic Science International: Synergy
    Citation Excerpt :

    This issue pushed the research to develop many analytical tools to allow forensic characterisation of such products [3–5]. Drugs for erectile dysfunction such as phosphodiesterase type 5 (PDE5) inhibitor medications belong to a special class of illegal pharmaceutical products sold on the Internet [6–8]. They are the most commonly counterfeited medicines in Europe and they were also found in dietary supplements [9,10].

  • Packaging analysis of counterfeit medicines

    2018, Forensic Science International
    Citation Excerpt :

    The authentication of the boxes, blisters, vials, syringes, and leaflets parts of the packaging could also be completed with analytical techniques usually used for the chemical investigation of the drug product. While many papers have been published about the analysis of the bulk medicines themselves [8–14], very few have focussed on the authentication of the primary or secondary packaging of suspected counterfeits. Rodomonte et al. studied the colour of the boxes of counterfeits with a handheld colorimeter [15].

  • ATR-FTIR characterization of generic brand-named and counterfeit sildenafil- and tadalafil-based tablets found on the Brazilian market

    2017, Science and Justice
    Citation Excerpt :

    In this work, ATR-FTIR was used to characterize a variety of genuine, both generic and brand-named, tablets based on PDE5 inhibitors sildenafil and tadalafil, legally produced and marketed in Brazil. Although several studies based on infrared profiling [12,14-20] have been published over this decade concerning identification of counterfeit samples of Viagra and Cialis, most of these works do not consider nor discuss the compositional characteristics of the samples, employing spectral profiling mostly as a data source for chemometrical analysis, an approach which, although very powerful and useful, is usually beyond practical use in most forensic laboratories. The present approach was based on more direct and comparative practical analyses of the spectral profiles of tablet cores, applied not only to identify the AI of each tablet, but also the major excipients/adjuvants present.

View all citing articles on Scopus
View full text