FormalPara Key Points

Of 40 included drugs, almost all the ionisable and non-ionisable drugs excreted into saliva with saliva-to-plasma ratios ranging from 0 to 12.8. Ionisation state, protein binding, lipophilicity, polar surface area and hydrogen bond donor count influenced the excretion of drugs into saliva.

Despite expecting physicochemical properties to predict saliva excretion, a complex interplay of other factors, possibly physiological variations and active transport may also play a role. This emphasises the need for further research and considerations in future clinical trials.

Concerning the suitability for saliva-based therapeutic drug monitoring (TDM), 75% of the commonly used drugs were considered likely or possibly suitable due to high saliva–plasma correlations and/or relatively low variability in saliva-to-plasma ratios.

1 Introduction

Therapeutic drug monitoring (TDM) predominantly relies on plasma-based measurements to assess the relation between drug exposure and efficacy/safety. However, the invasiveness of sampling blood presents several challenges, particularly in outpatient settings. It is painful and can be uncomfortable, which can lead to patient reluctance and non-compliance, particularly in settings where frequent monitoring is required. Collecting plasma samples, then transporting and analysing them necessitates advanced and expensive laboratory equipment, trained personnel and quality control measures. Further, pathology services supporting TDM are predominantly limited to tertiary hospitals or large commercial laboratories, rendering them out of reach for many patients. This centralised service model, which requires the transportation of samples, frequently results in lengthy delays before test results are available [1]. These delays hamper timely decision making, frustrating the pursuit of effective personalised drug dosing and dissuading healthcare professionals from embracing this approach. These difficulties can impede the successful implementation of TDM, particularly in outpatient settings, highlighting the need for alternative, more accessible and patient-friendly monitoring methods. These issues limit the uptake of TDM. Therefore, the potential benefits of TDM have not been realised [2].

Saliva is a patient-friendly matrix as it is less invasive and more comfortable to collect than plasma, making it more suitable for routine monitoring [3]. Another advantage is that, when compared with plasma, the drug concentration in saliva is considered the unbound proportion of the drug which is pharmacologically active [3]. Although saliva-based TDM has been studied for over 4 decades, routine clinical adoption remains uncommon [4]. Several studies have assessed the potential clinical utility of saliva-based TDM for various therapeutic classes such as anti-psychotics, anti-microbials, immunosuppressants and anti-thrombotic drugs. Encouragingly, some drugs [5,6,7] have consistently good correlations between plasma and saliva concentrations, although others do not [8,9,10]. These differences fuel the ongoing debate about the feasibility of saliva-based TDM.

Saliva testing has been commercialised for detecting drugs of abuse such as opioids and cannabis [11, 12]. For therapeutic drugs, saliva has been used to improve treatment of rifampicin-resistant tuberculosis with levofloxacin in Tanzania, a low-resource setting [13]. Levofloxacin concentrations in the saliva were measured [14] and used to quickly identify patients subtherapeutic levofloxacin concentrations that may have otherwise remained undetected and increased the risk of develo** resistance. This finding underscores the potential significant health benefits associated with saliva-based TDM. However, the broader translation of this approach into the clinic is impeded by a lack of understanding of the relationship between salivary and systemic drug concentrations. There is a crucial need for drug disposition studies in saliva to address this gap. When designing such studies, it is desirable to consider drug properties that govern the extent of saliva excretion of a specific drug. While physicochemical properties are commonly believed to predominantly influence excretion into saliva, the precise real-life mechanism remains unclear.

This study aimed to retrieve data on saliva-plasma concentration and subsequently determine the physicochemical properties that influence the excretion of drugs into saliva to increase the foundational knowledge to underpin saliva-based TDM.

2 Methods

2.1 Systematic Literature Review

The study protocol was registered with PROSPERO (CRD42023392728) and findings are reported according to the Preferred Reporting Items for Systematic Review and Meta-analysis of Individual Participant Data (PRISMA–IPD) guidelines (Supplementary Table S1) [15].

2.1.1 Information Sources and Search Strategy

Medline, Web of Science and Embase databases were searched for relevant literature. The searches included literature published between January 1974 and January 2023. Each database was last searched or consulted in June 2023. Search terms included (human OR man OR patients OR subjects) AND (medication OR drug) AND (saliva OR oral fluid) AND (plasma OR serum OR saliva-plasma ratio OR saliva-serum ratio) AND (pharmacokinetic* OR drug monitoring OR TDM OR penetration OR distribution OR concentration*). Topic search (TS), title (TI) and abstract (AB) search were used for Web of Science, whereas title, abstract and Medical Subject Headings (MeSH) terms were used for Medline. The search strategy for each database is presented in supplementary material. Advanced search strategies were used for all searches conducted. The database searches were limited to only human studies. Hand searching of references in articles was used to identify potential additional studies.

All searches were conducted by one author (TN) and independently checked by RC. Studies from these databases were imported into Covidence (Covidence systematic review software, Melbourne, Australia) which was used to manage all records exported from the databases and registry searches. Any duplicate reports were removed.

2.1.2 Selection Process

Records were screened based on the relevance of the study title and abstracts to the research question. Studies that met the inclusion criteria were subjected to full-text review and assessed against the inclusion and exclusion criteria.

Studies were included if they: (1) included ≥ 5 paired samples collected in both saliva and plasma/serum/blood per dose interval per subject; (2) included ≥ 10 subjects to capture the variability, in cases where the drug did not excrete into saliva, ≥ 5 subjects were considered sufficient; (3) quantitative methods were used to determine drug concentrations and (4) were prospective clinical trials. Studies were excluded if the drug studied can undergo biotransformation in salivary glands.

Two reviewers (TN and RC) independently screened each record retrieved based on the study title, abstract and the full text of eligible reports. If required, any disagreements were resolved by referral to a third independent reviewer (SS). There were no records screened or assessed by full text that required translation into the English language to determine eligibility.

2.1.3 Data Collection Process

An electronic data extraction form in Microsoft Excel was created to collate all relevant information from each eligible study. Saliva and plasma pharmacokinetic data including concentration-time data, peak concentration (Cmax), area under the concentration–time curve from 0 to the last timepoint (AUC0–t), saliva-to-plasma ratio, correlation coefficient (R) between saliva and plasma AUC, sampling method, sampling time (at steady state or not), saliva pH, saliva flow, analytical method and participant characteristics, were extracted. Corresponding authors were contacted to provide clarification of any missing or unclear information. If authors were uncontactable or did not respond, the information was recorded as not available. Two independent reviewers (TN and RC) captured the data. If needed, any disagreements were resolved by referral to a third independent reviewer (SS).

2.1.4 Risk of Bias Assessment

To assess the risk of bias in the studies, a checklist (Supplementary Table S2) was developed based on the risk of bias in non-randomised studies (ROBINS-I) tool [16] and relevant criteria regarding pharmacokinetic studies as proposed by Gafar et al. [17]. The tool addressed five domains: potential confounders, measurement of outcomes, study design and experimental methods, saliva sampling procedure and sample storage conditions. The maximum points obtained from this checklist was 12. The overall risk of bias for each study was determined as high (9–12 points), moderate (5–8 points) and low (≤ 4 points) quality. Two reviewers (TN, RC) independently applied the tool to each study. Any discrepancies in assessment of the risks of bias were resolved by referral to an independent reviewer (SS) until consensus was reached.

2.2 Data Analysis

2.2.1 Pharmacokinetic Studies in Saliva and Plasma

The primary pharmacokinetic measures were AUC and saliva-to-plasma ratio which were directly extracted from the studies. If the saliva-to-plasma ratio was unavailable, it was calculated by dividing the AUC in saliva by the total plasma/serum AUC. To enable analyses across studies, studies reporting mean AUC values based on unbound plasma/serum concentrations were converted to total concentrations using published protein-binding data. If the AUC was not reported, it was calculated based on the trapezoidal rule with concentration–time data extracted from figures using WebPlotDigitizer (available at https://automeris.io/WebPlotDigitizer/). Where multiple studies had assessed the saliva-to-plasma ratio of the same drug, the arithmetic mean was calculated.

To assess the utility of saliva-based TDM across the studies, correlation coefficients (R), describing the relationship between saliva and plasma drug exposure, were compared across studies. If not reported, they were computed from the concentration–time curves obtained in saliva and plasma. When available, the variability of saliva-to-plasma ratios between and within studies was assessed. Two criteria for suitability for saliva-based TDM were adequate saliva-plasma correlation (R2 > 0.6) and relatively low variability (coefficient of variation [CV%] < 20%) in saliva-to-plasma ratios. These arbitrary limits are based on statistical classification [18] and the expected purpose of saliva testing as a semi-quantification of drug concentration. Drugs were classified as follows: (1) likely suitable if they had both criteria, (2) possibly suitable if they had either one of the two, (3) unlikely if they did not meet either of the two criteria or were not detected in saliva and (4) unclear if there was a lack of pharmacokinetic data. All statistical analyses were performed using the software R (R Foundation for Statistical Computing, Vienna, Austria).

2.2.2 Determination of Physicochemical Properties

PubChem [19] and Drugbank [20] were searched for drug properties including pKa (negative logarithm of the ionisation constant of an acid), logP (lipid solubility), molecular weight, physiological charge, hydrogen bond donor count, hydrogen bond acceptor count, polar surface area, rotatable bond count and fraction of drug unbound to plasma protein. If these data were unavailable, the parameters were calculated based on quantum mechanics modelling using the Epik functionality in Maestro (Schrödinger, LLC, New York, NY, 2021). The drugs were classified into four groups based on their ionisability: acidic, basic, neutral or amphoteric (Supplementary Table S3).

2.2.3 Physicochemical Properties Predicting Saliva Excretion

For ionisable (acidic, basic and amphoteric) drugs, the degree of ionisation demonstrated by the Henderson–Hasselbalch principle and fraction unbound were hypothesised to be the most influential factors to drive saliva excretion. This is because most drugs are weakly acidic or basic and hence may be ionised at physiological pH; and theoretically only the free fraction (not bound by plasma protein) of a drug can cross membranes. Thus, theoretical salivatotal-concentration/plasmatotal-concentration ratios based on this principle with adjustment for fraction unbound [21] were calculated and compared with the observed ratios (Eqs. 1 and 2). In which, pHs and pHp denotes the average pH of saliva and plasma (6.7 and 7.4, respectively) [22]. Saliva pH was varied from 5.8 to 7.6 to mimic patho-physiological variation [23, 24].

$${\text{For an acid:}}\quad {\text{Saliva - to - plasma}}\;{\text{ratio}} = \frac{{1 + 10^{{\left( {{\text{pH}}_{{\text{s}}} - {\text{p}}K_{{\text{a}}} } \right)}} }}{{1 + 10^{{\left( {{\text{pH}}_{{\text{p}}} - {\text{p}}K_{{\text{a}}} } \right)}} }} \times \frac{{f_{{\text{p}}} }}{{f_{{\text{s}}} }}$$
(1)
$${\text{For a base:}}\quad {\text{Saliva - to - plasma}}\;{\text{ratio}} = \frac{{1 + {10}^{{\left( {{\text{p}}K_{{\text{a}}} - {\text{pH}}_{{\text{s}}} } \right)}} }}{{1 + {10}^{{\left( {{\text{p}}K_{{\text{a}}} - {\text{pH}}_{{\text{p}}} } \right)}} }} \times \frac{{f_{{\text{p}}} }}{{f_{{\text{s}}} }},$$
(2)

where fs is fraction unbound drug in saliva, fp is fraction unbound drug in plasma, pHs is saliva pH and pHp is plasma pH.

Associations between physicochemical properties and saliva excretion were assessed by non-parametric Spearman rank correlation analyses (package ‘PerformanceAnalytics’: Econometric Tools for Performance and Risk Analysis [25]). The Spearman method was selected because distributions of all variables were not normal, variability of the outcome–predictor relationship was not constant and the relationship was not linear. Logarithmic transformation was performed when no significant correlations were observed on a linear scale. Stratification into low and high excretion with arbitrarily defined cut-off values of the median saliva-to-plasma ratio was performed. A p < 0.05 was considered statistically significant. To predict the likelihood of a specific drug excreting into saliva (saliva-to-plasma ratio \(\ge\) 0.1), a framework was developed based on the results of correlation analyses.

3 Results

3.1 Systematic Literature Review

A total of 1649 articles were identified (Fig. 1). Following the removal of duplicates, 1586 studies were screened based on title and abstract. A total of 197 articles were assessed for eligibility based on the full text review. Of these, 42 studies were included. A total of 155 articles were excluded because (1) the sample size was inadequate, (2) sampling was too sparse to assess AUC, (3) studies were conducted in animals, (4) no saliva or plasma concentration–time profiles were reported or (5) compounds that were neither exogenous nor therapeutic (e.g. endogenous compounds or illicit drugs). One study on carboplatin was excluded because carboplatin acts as a prodrug and aquates into a variety of daughter drugs in aqueous solution [26] and thus the properties of its parent cannot be predicted. Prednisone (prodrug) and prednisolone (active metabolite) were excluded from the initial dataset because prednisone can be converted to prednisolone by an enzyme found in the salivary glands—11 beta-hydroxysteroid dehydrogenase 2 (11-BHSD2) [27]. As such, saliva excretion of this parent–metabolite pair was considered unpredictable.

Fig. 1
figure 1

PRISMA diagram for the systematic review

Full size image

3.1.1 Study Characteristics

More than half of the studies (64.3%, 27/42) were conducted in healthy subjects and the median number of participants was 19 [interquartile range (IQR) 12–24] (Table 1). Overall, 40 drugs were evaluated including anti-microbials, anti-arrhythmics, analgesics, anti-cancer agents, anti-psychotics and immunosuppressive drugs [28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70]. Half of the studies (52.4%, 22/42) included collection of samples after a single dose, with 40.5% (17/42) of the studies assessing saliva excretion at steady-state.

Table 1 Clinical pharmacokinetic studies included in the study (grouped by acidity/basicity)
Full size table

Two main methods were used to collect saliva including spitting or draining of saliva into a container (unstimulated) and stimulating saliva flow with parafilm, gum, acids or absorbent cotton swabs such as Salivette (stimulated). About half of the studies (52.4%, 22/42) used the unstimulated method with most of the remaining studies (85%, 17/20) using the stimulated method. One study did not report the collection method and two studies used both collection methods to compare their results. The study that compared three different stimulated methods of citric acid stimulation, citric acid-impregnated Salivette device and a neutral Salivette showed no significant differences in saliva codeine concentrations [39]. Using the unstimulated method after mouth rinsing improved saliva–plasma correlation for tacrolimus when compared with the stimulated method [68]. Retention of the drug formulation in the mouth of participants after oral intake affected the saliva concentration of codeine [39]. Abnormally high codeine concentrations in the first hours after liquid drug administration were reported [39]. Excess of the upper limit of quantification for a tacrolimus assay was associated with high transferrin (blood biomarker) concentrations, suggesting potential blood contamination in saliva samples [68]. Saliva pH varied across saliva collection methods with or without stimulation and rinsing. Generally, saliva stimulated with acids resulted in a lower pH (mean 3.8; range 2.5–6.2) than unstimulated saliva (7.3; 6.6–8.1).

3.2 Risk of Bias Assessment

All the studies had a low-to-moderate risk of bias (Supplementary Table S4). Potential bias included studies that were not conducted at steady state (n = 26), did not account for potential confounders (n = 7), did not measure saliva pH even though the drugs were sensitive to saliva pH changes (n = 13), did not provide information about storage conditions for drugs unstable at room temperature (n = 2) and/or the analytical procedures were not described adequately (n = 15).

3.3 Pharmacokinetic Studies in Saliva and Plasma

All the drugs studied excreted with saliva, except for amikacin. Most studies (71.4%, 30/42) reported correlation coefficients (R) of saliva and (unbound/total) plasma AUC with a median (IQR) of 0.93 (0.73–0.97) (Table 1). The correlations in some studies were dependent on dose (codeine, cocaine), time since drug administration (metronidazole, codeine) and saliva collection methods (indinavir). For the drugs with time-dependent correlations, an improvement in the correlation was reported after the absorption phase was completed, possibly due to oral contamination. Poor correlations could be improved by measuring unbound plasma drug concentrations instead of total plasma concentrations for disopyramide [47]. Correlations between tacrolimus concentrations in saliva and plasma were improved by rinsing the mouth before saliva collection [R = 0.50, p = 0.02 (no rinsing) versus R = 0.71, p = 0.01 (with rinsing)] [39].

Many studies did not report variability between subjects in saliva-to-plasma AUC ratios. Only 15 studies (35.7%) reported between-subject saliva-to-plasma ratio (unbound/total plasma) variability with a median (IQR) CV% of 17.5% (10.7–28.5%). Between-subject saliva-to-total plasma concentration variability was substantially higher than saliva-to-unbound plasma variability for disopyramide [47]. Six drugs were investigated in more than one study. The variability (CV%) in saliva-to-plasma ratios between studies for five drugs (gatifloxacin, disopyramide, codeine, moxifloxacin, voriconazole) was less than 20% and a maximum of 26.5% for metronidazole.

Only five studies assessed the impact of saliva pH on drug concentrations in saliva. Of these, three were conducted in healthy subjects and two in patients. Saliva pH was more acidic and variable in patients (mean 5.4; range 2.5–8.1) than in healthy subjects (mean 7.1; range 6.2–8.2). The saliva-to-plasma ratios for codeine decreased with increasing pH of saliva [40]. The saliva collection method is thought to alter its pH, so its effect on drug concentrations in saliva was considered. Saliva codeine concentrations collected with acid candy stimulation were higher than with either acid-impregnated or neutral cotton swabs at early collection times; however, these differences were not statistically significant [39].

Only one study determined whether pathophysiological conditions influenced salivary excretion. In that study, indinavir concentrations were higher and more variable in people living with human immunodeficiency virus (HIV) than in healthy subjects (p < 0.01) [36]. Indinavir pharmacokinetics may be influenced by many factors such as disease-related alterations in physiology, potential drug interactions from anti-retroviral therapy and patient-related variables such as demographics or drug adherence [71,72,73].

Concerning the suitability of saliva-based TDM of included studies, 11 drugs (27.5%) were likely suitable. Nineteen drugs (47.5%) were determined to be possibly suitable. Three drugs (7.5%) were determined to be unlikely to be suitable due to highly variable saliva-to-plasma ratios and it was unclear whether six drugs (15.0%) are suitable due to a lack of sufficient pharmacokinetic (PK) data to be evaluated. Amikacin was considered unlikely to be suitable because it was undetected in saliva. Details of specific drugs are presented in Table 2.

Table 2 Suitability of saliva-based TDM of included studies
Full size table

3.4 Physicochemical Properties Predicting Saliva Excretion

Among the 40 selected drugs, there were 36 ionisable drugs (21 basic, 10 amphoteric and 5 acidic) and 4 un-ionisable (neutral) drugs (Supplemental Table S3). In the basic and amphoteric groups of drugs, eight extremely weak acids with pKa > 12 or extremely weak bases with pKa < 3 (3 amphoteric and 5 basic) were also included in the neutral group. The physicochemical properties of these drugs are presented in Supplementary Table S5. Most of the drugs were excreted with saliva with saliva-to-plasma ratios ranging from 0 to 12.8 (Table 3). Nine drugs (22.5%) showed limited excretion (saliva-to-plasma < 0.1). Nine drugs (22.5%) showed saliva accumulation (saliva-to-plasma ratios > 1). The median saliva-to-plasma ratios in the amphoteric, basic, and acidic groups were 2.8-fold higher, 1.9-fold higher and twofold higher (Wilcoxon rank sum test, all p > 0.05), respectively, when compared with the neutral group of drugs.

Table 3 The observed saliva-to-plasma ratio per drug group
Full size table

Theoretical saliva-to-plasma ratios calculated based on the Henderson–Hasselbalch principle are presented in Fig. 2. Results of Spearman correlation analyses are summarised in Fig. 3 and detailed in Supplementary Figs. S1–S5.

Fig. 2
figure 2

Observed versus predicted saliva-to-plasma ratios by applying Henderson–Hasselbalch principle with adjustment for protein binding for acid (A), basic (B), amphoteric (acid; C) and amphoteric (base; D) groups, respectively. Observed saliva-to-plasma ratios (grey circles) and line of best fit (dashed black line) using linear regression analysis with 95% confidence band around the line (grey area). Identity line (solid black line; x = y)

Full size image
Fig. 3
figure 3

Correlations between physicochemical properties and saliva excretion for basic (A), amphoteric (B, C) and neutral (D–F) drugs. Spearman correlation scatter plot [linear regression (dash black line) with its confidence interval (grey area)] for saliva-to-plasma (S/P) ratios versus significant physicochemical properties (p < 0.05). Upper left corner with R, Spearman correlation coefficient and p, corresponding p value. See Supplementary Figs. S1–5 for correlations between all of the investigated properties and saliva excretion

Full size image

For the acidic group drugs, the state of ionisation and fraction unbound could explain drug excretion into saliva (via Henderson–Hasselbalch principle) due to the relatively good correlation between observed and predicted saliva-to-plasma ratios (Fig. 2) (R2 = 0.85, p = 0.02, n = 5).

For the basic group drugs, low ionisation and high fraction unbound were associated with higher saliva excretion of nine weakly basic drugs (pKa < 9) namely clarithromycin, clonazepam, eletriptan, indinavir, isoniazid, lamotrigine, maraviroc, piroxicam and risperidone. Stronger bases did not follow the Henderson–Hasselbalch principle.

Spearman correlation analysis showed that pKa was the only significant predictor of higher saliva excretion after logarithmic transformation (R = 0.53, p = 0.03, n = 17). A higher fraction unbound to plasma protein appeared to be associated with higher saliva excretion (R = 0.43, p = 0.08, n = 17). In the subgroup analysis of drugs with a higher saliva excretion (saliva-to-plasma ratios > 0.5, n = 12), higher excretion was associated with higher basicity (R = 0.73, p = 0.01) and a higher count of positive charge (R = 0.73, p = 0.01).

For the amphoteric group drugs, the Henderson–Hasselbalch principle did not predict saliva excretion (n = 10), but drugs with fewer hydrogen-bond donor counts and lower polar surface areas had greater saliva excretion (p = 0.01 and p = 0.03, respectively).

For the neutral or very weak basic/acidic group drugs, greater saliva excretion was associated with a higher fraction unbound to plasma protein (p = 0.004), lower lipophilicity (p = 0.04) and fewer hydrogen bond donor groups (p = 0.03).

On the basis of these identified properties and inspired by Lipinski’s rule of five for assessing druglikeness [74], a quantitative summary of drug properties was developed to describe the likelihood of drug excretion into saliva (Table 4). Initially, drugs capable of being excreted into saliva (with saliva-to-plasma ratio ≥ 0.1) and drugs with limited capability of being excreted into saliva (with saliva-to-plasma ratio < 0.1) were determined. Subsequently, physicochemical properties of those drugs were summarised and compared between those two groups (details provided in Supplementary Table S6). Lastly, the cut-off values for each drug property associated with higher excretion into saliva were determined and rounded to be multiples of six when appropriate due to their alignment with the rounded numerical data. It was assumed that passive diffusion was the primary mechanism for drug excretion into saliva, disregarding the role of transporters.

Table 4 Summary of drug properties to describe the high likelihood of excreting into saliva of drugs
Full size table

4 Discussion

We investigated the correlation of the physicochemical properties of a drug to its excretion into saliva. While conducting TDM using saliva is most applicable to drugs which currently already undergo TDM (e.g. tacrolimus, voriconazole), we opted not to limit our study solely to these drugs. This enables the potential future application of drug concentration monitoring (perhaps to monitor adherence to therapy) in saliva even for drugs for which TDM is currently not recommended, such as metformin. Saliva excretion could be described by ionisation state and protein binding for basic and acidic drugs, whereas polar surface area and hydrogen bond activities were important determinants for amphoteric drugs. Our study corroborates the findings of previous investigations, demonstrating consistency in the impact of ionisation [75]. Moreover, our larger study extends the current knowledge base by providing novel insights into the specific drug properties per the classification of ionisability, shedding light on previously ambiguous findings [10].

All the drugs excreted with saliva, except amikacin [45]. Amikacin has many hydrogen bonding groups making it highly hydrophilic and many positive charges which may hinder its permeability into saliva. The accumulation of basic drugs in saliva may reflect ion trap** where weak bases tend to accumulate in body compartments of relatively low pH and vice versa [76].

Ionisable drugs had greater saliva excretion compared with un-ionisable drugs. The equilibrium between ionised and un-ionised forms explains the passive transversal of the un-ionised form through lipid bilayer membranes [77]. Involvement of active transport of drugs into saliva also contributes. This is supported by over fivefold higher concentrations of drugs such as chloroquine and spiramycin in saliva when compared with plasma [30, 42]. In addition, large drug molecules, such as tacrolimus, excrete into saliva despite being unlikely candidates for passive diffusion [68, 78]. This suggests the involvement of active transport, such as BSEP (ABCB11), an efflux transporter expressed in salivary glands. Indeed, genetic variation in BSEP could predict the concentration of tacrolimus in saliva [68]. Finally, several other transporters including OCT1, OCT3, OCTN2, MRP1, MRP2, MATE1, BSEP and P-gp are expressed in salivary glands [79,80,81,82]. However, the contribution of drug transporters on saliva excretion has not been extensively studied and thus further research is needed.

Excretion into saliva of acidic and some weakly basic drugs followed the Henderson–Hasselbalch principle [83]. For strongly basic drugs, stronger basicity and higher positive charge were the significant predictors of higher saliva excretion. Again, this finding can be explained by the ionised and un-ionised equilibrium or the involvement of transporter-mediated mechanisms.

In the amphoteric group of drugs, saliva-to-plasma ratios were described by polar surface area and hydrogen bond donor count. Polar surface area, as an indicator of the accessibility of the drug to solvent by hydrogen bond formation, showed an inverse relationship with permeability across cell monolayers [84]. Consequently, polar surface area is linked to the transport of drugs across epithelial cells. Hydrogen bond donor count negatively impacted saliva-to-plasma ratios, aligning with prior findings that hydrogen bond donors reduce cell membrane permeability by increasing interactions with the lipid bilayer [85, 86]. For the neutral group of drugs, fraction unbound to plasma protein, lipophilicity and number of hydrogen bond donor groups could predict saliva excretion. If passive diffusion is assumed the main mechanism, a higher fraction unbound to plasma protein and lower hydrogen bond donors might explain greater drug excretion from plasma to saliva.

Together, the evidence suggests both active transport and passive diffusion contribute to drug excretion into saliva. This is consistent with drug excretion into other matrices such as breast milk, placenta and urine. In addition to transporters, drug excretion into breast milk depends on polar surface area, molecular weight and ionisation in plasma [83]. Drug excretion into placenta depends on transporters, lipophilicity, protein binding, drug ionisation, molecular weight and foetal–maternal clearance [87, 88]. Likewise, drug excretion into urine depends on transporters, lipophilicity, drug ionisation and molecular weight [89]. Overall, protein binding and drug ionisation are the two common factors that influence excretion into various biological compartments.

We might have underestimated saliva excretion by calculating saliva-to-total plasma ratios because of the limited data on protein binding in individual studies. However, the total and unbound drug concentrations in plasma are often correlated which results in marked correlations between total drug concentrations in plasma and saliva (Table 1). As such, the underestimated saliva excretion was unlikely to have impacted the results because of the use of consistent total plasma concentrations in the pooled dataset.

Physiological factors, namely saliva pH, might also influence saliva excretion of drugs. Theoretically, saliva pH would be most influential as based on the Henderson–Hasselbalch principle, at saliva pH from 6 to 8, for basic drugs with a pKa over 6 and acidic drugs with a pKa under 8 excretion into saliva is profoundly affected by minor variations in saliva pH (Supplementary Table S7) [90]. Although the drugs included in approximately half of the studies (22/42) are prone to saliva pH variation, only five of the studies controlled for this variable [36, 40, 47, 48, 68]. In these studies, three basic drugs with pKa over 6 (codeine, disopyramide and indinavir) exhibited decreasing saliva-to-plasma ratios with increasing saliva pH [36, 40]. Therefore, future clinical trials should consider the impact of saliva pH on saliva-to-plasma ratios for drugs ionised at (patho)physiological pH. Pathological factors can also alter saliva excretion. For example, indinavir saliva-to-plasma ratios were higher and more variable in HIV patients compared with healthy subjects (68% versus 35% (intra-individual variability) and 37% versus 8% (inter-individual variability), respectively [36].

Of the drugs potentially suitable for saliva-based TDM, plasma-based TDM is routinely performed for phenytoin, tacrolimus, voriconazole and lamotrigine [91]. Since these drugs are often used in outpatient settings, saliva-based TDM could facilitate patient-led care with support from community healthcare professionals. However, further research using well-designed studies and robust methodologies is required to better understand drug excretion into saliva and factors contributing to variability in this clearance pathway. The bottleneck for the clinical implementation of saliva-based TDM is the fear of inaccuracy of saliva concentration measurement and the variability in saliva-to-plasma ratios, mainly due to variability in saliva pH, saliva flow and oral contamination. To improve the measurement accuracy, technological advances in sample clean-up, analytical sensitivity (to minimise false-negative results) and specificity (to minimise interferences in saliva) as well as standardisation in saliva collection protocols are needed. Controversies exist with optimal saliva collection methods. One study favoured a stimulated method with acids due to an improvement in saliva–plasma correlation [36], while another study favoured an unstimulated method due to its more accurate saliva concentrations (without being significantly diluted) [68], whereas a third study found no significant differences between stimulated and unstimulated methods [39]. These findings highlight the importance of method validation and comparison to minimise potential interference(s) and identify the most suitable saliva collection method. Strategies for analytical method development and validation of saliva-based pharmacokinetic studies can be found in a recent opinion paper [92]. Most therapeutic thresholds for dose decision making are defined in plasma/serum. Once adequate pharmacokinetic data in plasma and saliva are available, population pharmacokinetic modelling can help predict plasma exposure from saliva, rendering blood collection unnecessary. For example, a saliva population pharmacokinetic model for clonazepam, an anti-epileptic drug, has been developed that can accurately predict steady‐state trough plasma concentrations from concentrations obtained in saliva [41]. Another approach that uses physiologically based pharmacokinetic modelling was able to describe variability in both saliva and plasma exposures of matrine, an anti-cancer drug [93]. These examples showcase the potential to facilitate a model-informed personalised dosing approach using drug concentrations obtained from saliva. Indeed, physiologically based pharmacokinetic models that predict drug exposure in placenta and breast milk have been used to guide prescribing and inform regulators during pregnancy and lactation [94,95,96]. A recent prospective pharmacokinetic and simulation study showed that gentamicin monitoring in newborns using saliva samples correctly guided 81% simulated dose regimens [97]. This suggests that saliva-based TDM can reliably inform optimal dosing. Further studies comparing the impact of drug concentrations obtained in saliva and plasma on dose decisions are required. In situations where collection of blood is challenging such as for children, in the outpatient setting or in limited resource settings, plasma sampling can be omitted if determination of drug concentrations in saliva has been adequately validated.

5 Conclusions

Saliva-based TDM is considered potentially suitable for phenytoin, tacrolimus, voriconazole and lamotrigine. For other drugs, further research is required to understand the physicochemical factors that influence the excretion of drugs into saliva as well as the influence of drug transporters and physiological factors. Saliva excretion is affected by saliva pH, plasma protein binding, lipophilicity, hydrogen bond donor count and polar surface area; therefore, consideration of these variables for saliva-based TDM is required prior to implementation into practice. Further translational research on these promising candidates will facilitate implementation of this person-centred TDM approach to enhance patient outcomes and quality of care.