Overall, 737 samples from 185 people were included in the analysis. A summary of the key characteristics for the people included in the study and the samples analyzed can be found in Tables 1 and 2, respectively.

Median concentrations (IQR) for people with HIV (PWH) with HIV-1 RNA < 20 copies/mL or ≥ 20 copies/mL were 1,480 µg/L (1,097; 1,955) or 1,180 µg/L (879; 1,570) for CAB (p = 0.001), and 77 µg/L (53; 107) or 63 µg/L (47; 87) for RPV (p = 0.001), respectively.

Median concentrations (IQR) for PWH with “target not detected” or “target detected” were 1,515 µg/L (1,130; 1,955) or 1,290 µg/L (939; 1,760) for CAB (p < 0.001), and 78 µg/L (54; 111) or 67 µg/L (48; 93) for RPV (p = 0.003), respectively. Association of drug concentrations with viremic episodes.

Using 800 µg/L for CAB and 38 µg/L for RPV as thresholds to define low concentrations, 46 (6.2%) samples had low concentrations of CAB, 47 (6.4%) samples had low concentrations of RPV, and 17 (2.3%) had low concentrations for both drugs. 11/64 samples (17.5% [9.4; 29.5]) with low CAB concentrations and 74/677 samples (10.9% [8.7; 13.6]) with normal CAB concentrations had HIV-1 RNA ≥ 20 copies/mL, respectively. 12/64 samples (18.8% [10.5; 30.8]) with low RPV concentrations and 73/676 samples (10.8% [8.6; 13.4]) with normal RPV concentrations had HIV-1 RNA ≥ 20 copies/mL, respectively. Low concentrations for CAB and RPV were associated with odds ratios of 1.7 (0.8; 3.5) and 1.9 (0.9; 3.8) for HIV-1 RNA ≥ 20 copies/mL, respectively. Thresholds found by maximizing the sums of sensitivity and specificity were 1,240 µg/L and 76 µg/L for CAB and RPV (Fig. 1), respectively. Using these thresholds, 85 (11.5%) samples had low levels of CAB, 187 (25.4%) had low levels for RPV, and 190 (25.8%) had low levels for both. 48/275 samples (17.5% [13.3; 22.6]) with low CAB concentrations and 37/465 samples (8.0% [5.7; 10.9]) with normal CAB concentrations had an HIV-1 RNA ≥ 20 copies/mL, respectively. 58/377 samples (15.4% [12.0; 19.5]) with low RPV concentrations and 27/363 samples (7.4% [5.0; 10.8]) with normal RPV concentrations had a HIV-1 RNA ≥ 20 copies/mL, respectively. Low concentrations of CAB or RPV were associated with odds ratios of 2.4 (1.5; 4.0) and 2.3 (1.4; 3.8) for HIV-1 RNA ≥ 20 copies/mL, respectively.

Receiver operating characteristics (ROC) for the sensitivity and 1-specificity throughout the range of possible ‘optimized’ thresholds for Cabotegravir (top) and Rilpivirine (bottom) on the left. The results of the maximized sum of sensitivity and specificity can be found on top of the figure. For both drugs, the right graphs depict the sums for sensitivity and specificity for a range of drug concentrations

Figure 2 depicts the division of all pairs of measurements with HIV-1 RNA ≥ 20 copies/mL into four quadrants, of which the left lower quadrant is the one for blood samples with low concentrations for both CAB and RPV.

Scatter plot of measurement pairs for cabotegravir (CAB) and rilpivirine (RPV) concentrations for all samples with HIV-1 RNA ≥ 20 copies/mL (top) and ≥ 50 copies/mL (bottom). Red dashed lines indicate the data-derived thresholds for CAB and RPV, tables in the upper right corner of the plots indicate the number of samples per quadrant in relation to the number of samples with HIV-1 RNA ≥ 20 copies/mL (n = 87) or HIV-1 RNA ≥ 50 copies/mL (n = 18), respectively. Colors indicate the result of HIV-1 RNA quantification, also indicated by the numbers next to each measurement

In our study we found an association between drug concentrations of CAB and RPV and HIV-1 RNA measured in the same sample. Samples with HIV-1 RNA ≥ 20 copies/mL had on average lower concentrations for both CAB and RPV (Fig. 3). Likewise, when applying the thresholds of 800 µg/L for CAB and 35 µg/L for RPV, a higher proportion of samples with low drug concentrations had HIV-1 RNA ≥ 20 copies/mL. From a cohort in Spain, similar data were reported, where in four people with HIV-1 RNA ≥ 50 copies/mL lower concentrations of both drugs were found [5].

Box- and jitter-plots of the concentrations of Cabotegravir and Rilpivirine by status of virologic suppression. Numbers below the plots indicate the median together with the interquartile ranges for the respective drug

It should be noted that the trough concentrations for CAB (1,445 µg/L [1,064.8; 1921.2]) and RPV 75 µg/L (51.0; 105.0) found in our study were at the lower end of the range reported in clinical trials, ranging from roughly 1,500 to 1,800 µg/L and 70–100 µg/L [9, 10, 11], respectively. At least one other real-world study also reported lower drug levels than in clinical trials [6].

Since the optimal thresholds for CAB and RPV in the context of therapeutic drug monitoring (TDM) remain a topic of debate, we explored alternative cut-offs by applying ROC analyses, selecting the concentration that maximized the sum of sensitivity and specificity for the binary outcome (HIV-1 RNA < 20 copies/mL: yes vs. no). This led to thresholds of 1,240 µg/L and 76 µg/L for CAB and RPV, respectively. While both values are considerably higher than the ones used before, they are still lower than the ones reported from four-weekly dosing [11]. It has been advocated in models for oral RPV before that often-used thresholds might be too low and concentrations of up to 100 µg/L were discussed, however, including people without prior treatment [4]. Interestingly, a threshold of 100 µg/L was comparable in terms of sensitivity and specificity to 76 µg/L in our ROC analysis, leading to an only slightly lower sum of the two test performance characteristics (Fig. 1). It is noteworthy that the AUCs of both ROC analyses are quite low; however, it must be kept in mind that the contribution of both drugs might be relevant and that therefore the AUC for each single drug might not be indicative of the discriminatory ability of the combined use.

In this context, it should be noted that a trend toward higher efficacy with four-weekly compared to eight-weekly dosing has been observed [3], and that people with HIV (PWH) who were stable on CAB & RPV Q4W experienced virologic failure (soon) after switching to Q8W [10]. Both phenomena are most easily explained by the differences in pharmacokinetics, as all other risk factors should contribute equally in both settings.

When visualizing paired CAB and RPV measurements together with the data-derived thresholds for samples with HIV-1 RNA ≥ 20 copies/mL (Fig. 2), most samples fell into groups with low levels of at least one of the drugs. The highest detected viral loads were all found in the quadrant with low concentrations of both drugs. This quadrant also contained the highest proportion of samples with HIV-1 RNA ≥ 20 copies/mL. In contrast, only about 20% of samples with HIV-1 RNA ≥ 20 copies/mL had sufficient concentrations of both CAB and RPV. However, this also suggests that trough concentrations alone may not be sufficient to predict viremic episodes in PWH, a finding that has been noted previously [5].

Our study has several limitations. First, it is a retrospective study, meaning data collection occurred post hoc. However, the study site had implemented therapeutic drug monitoring for all individuals receiving CAB and RPV, ensuring that the primary outcome measure was assessed as a standard during virtually all visits. Second, our results are only based on trough levels, that don’t reflect the entire treatment period between two injections. However, they might be seen as a (good) surrogate, reflecting a minimum concentration during this time. Third, it is often argued that the Roche assay has a high variability compared to other tests in the field and therefore the results have to be interpreted with care. However, the following consideration is noteworthy in the context of our study: a higher variability adds more “noise” to the data, which makes it more difficult to detect a true signal. Finding such a signal is therefore even more likely to be “true”. This is particularly true as in the suppressed setting the variability will (almost) always be directed towards “false-high” viral loads, meaning that more often a detectable viral load will be found where there might be none in another assay. It should therefore be expected that more “false-high” HIV-1 RNA levels are detected, which is in favor of the null hypothesis and should therefore increase the type 2 error, instead of the type 1 error. In the same context it must be mentioned that low-level viremic events are not comparable to virologic failure. Due to the rare occurrence of virologic failure (none in this study sample), we had to focus on viremic events. However, for this reason we tried to validate our findings by applying our findings to the reported VFs from clinical trials and real-world data and found high rates of insufficient drug-levels. This still does not prove causality. Finally, all data were derived from a single clinical site, which may impact the generalizability of our findings. This is particularly relevant for the alternative drug concentration thresholds, as they may have been “overfitted” to the available data. From a clinical perspective, it is noteworthy that no virologic failure occurred among the individuals included in this study. This raises the question of whether low CAB and RPV concentrations are merely associated with viral blips and low-level viremia or if they serve as true predictors or risk factors for virologic failure. Low-level viremic events are not consistently linked to eventual virologic failure. However, when applying our thresholds to PWH with virologic failure in clinical trials (ATLAS-2 M, SOLAR) [1, 12] and real-world cohorts [5, 6, 13, 14, 15, 16, 17, 18, 19] where trough levels at the time of VF were reported, we found that in 25/28 cases (89.3%) low concentrations were present. After removing people with subtype A6 or major NNRTI resistance at baseline, 17/19 subjects (89.5%) and 18/19 subjects (94.7%) were low in either concentration for the 76 µg/L and 100 µg/L thresholds for RPV, respectively. For comparison: the expected proportion based on our sample would be 62.6% (59.0-66.1; p = 0.021 for a one-sided test for proportions, data not shown) and 78.2% (75.0-81.1; p = 0.086 for a one-sided test for proportions, data not shown) for the lower and higher thresholds of RPV, respectively. 9/18 (50%) people experiencing virologic failure had low levels for CAB and RPV as compared to an expectation of 27.7% based on the findings in our study sample (p = 0.021 for a one-sided test for proportions, data not shown). Lastly, it is likely that other factors like the size of the reservoir or intercurrent infections amongst others, could contribute to the occurrence of viremic episodes which has not been considered in the current analysis.