# Combinations of Artemisinin and Quinine for Uncomplicated Falciparum Malaria: Efficacy and Pharmacodynamics

## ABSTRACT

*P*< 0.001). Recrudescence was associated with shorter parasite clearance time (PCT) and longer treatment after the blood smear had become negative (eradication time). However, classification of patients to outcome—recrudescence or radical cure—was correct in only 77% of patients. The population kinetics of the parasitemia was estimated with nonlinear mixed-effect models. Several models were tested, but the best model was a monoexponential decline of the parasitemia in which the mean parasite elimination half-life was shorter after artemisinin (5.1 h; 95% confidence interval [CI], 4.9 to 5.2 h) than after quinine (8.0 h [95% CI, 7.5 to 8.3 h]). Attempts to simulate the initial increase of the parasitemia did not result in better models with a biologically plausible interpretation. Recrudescence was associated with slower parasite clearance and a higher simulated terminal parasitemia (

*P*

_{term}). The classification of patients to outcome groups based on

*P*

_{term}was correct in 78% of patients. The data suggest that parasite strains with reduced sensitivity to quinine are prevalent in Vietnam, with slower parasite clearance and consequent recrudescence. A single dose of artemisinin induces rapid parasite reduction and lowers the value of

*P*

_{term}, but to prevent recrudescence, this should be followed by quinine for at least 3 days after parasite clearance, or 5 days in total.

## MATERIALS AND METHODS

### Treatment.

### Clinical assessments.

### Population kinetics of the parasitemia.

*P*(0)], the duration of effective drug treatment, and the elimination rate converge in a single value: the parasitemia at the end of the treatment (

*P*

_{term}). The end of therapy was defined as the time of the last dose of quinine plus 8 h, which represents the duration of one dosing interval.

*P*

_{term}can be derived from the kinetic models of the parasitemia.

*P*

_{term}and the parasite density of the recrudescence on a semilogarithmic plot. This value is a rather crude estimate, connecting a simulated value and a single data point. However, it may serve as a reference to other literature.

### Statistical analysis.

^{2}tests with continuity correction for categorical parameters and with analysis of variance (ANOVA) or nonparametric tests for numerical parameters. The effects of treatment and kinetic parameters on the occurrence of parasite clearance or recrudescence were analyzed by logistic regression and in a Cox proportional hazard model. Statistical significance was accepted at P < 0.05.

## RESULTS

### Clinical assessment.

Characteristic | Result with regimen^{a} | ||
---|---|---|---|

Quinine (Q) | Artemisinin + 3 days of quinine (AQ3) | Artemisinin + 5 days of quinine (AQ5) | |

Sex (no. male/no. female) | 70/14 | 76/20 | 70/18 |

Median age (yr) [range] | 26.0 [7–60] | 26.0 [7–64] | 26.0 [7–60] |

Mean wt (kg) | 47.9 (46.1–49.7) | 46.3 (44.3–48.3) | 48.3 (45.9–50.7) |

Mean temp (°C) | 38.6 (38.5–38.7) | 38.6 (38.5–38.7) | 38.7 (38.6–38.9) |

Geometric mean P(0) (per μl) | 16,157 (12,642–20,646) | 16,123 (12,611–20,611) | 23,202 (17,888–30,091) |

^{a}

Patient outcome | No. (%) of patients with result | |||
---|---|---|---|---|

Quinine (Q) | Artemisinin + 3 days of quinine (AQ3) | Artemisinin + 5 days of quinine (AQ5) | Total | |

Total | 84 | 96 | 88 | 268 |

Early dropout, not evaluable | 3 | 1 | 1 | 5 |

Lost before day 7 | 2 | 3 | 5 | |

Failure | ||||

Early | 1 | |||

Late | 1^{a} | |||

Radical cure | 56 | 46 | 66 | 168 |

Recrudescence^{b} | 11 (16) | 28 (38) | 12 (15) | 51 (30) |

Early | 3 | 11 | 14 | |

Late | 8 | 17 | 12 | 37 |

Lost between days 7 and 28 | 10 | 18 | 9 | 37 |

^{a}

^{b}

*P*< 0.001 (χ

^{2}test). The 95% CIs of the differences are as follows: Q versus AQ3, 8 to 36%; Q versus AQ5, −11 to 13%; and AQ3 versus AQ5, 9 to 37%.

*P*(0), PCT, and ET for treatment groups and outcome are shown in Table 3.

### Kinetic models.

*t*

_{1/2el}) in groups Q7, AQ3, and AQ5 were 8.0 h (95% CI, 7.5 to 8.3 h), 4.8 h (95% CI, 4.6 to 5.0 h), and 5.3 h (95% CI, 5.2 to 5.5 h), respectively. The model estimates and derivatives, specified for treatment and outcome, are shown in Table 3. Instead of the estimated initial parasitemia (expressed as

*A*in the formulas), lag time (

*t*

_{lag}) is presented. This was calculated according to formula 4 in the Appendix. The elimination rate has been recalculated in response to

*t*

_{1/2el}. As a reference value for the observed data, the PCT was calculated from the kinetic estimates (PCT

_{calc}), setting the detection limit at 6 parasites/μl, the lowest parasitemia observed in this study. The kinetic models yielded the estimates of

*P*

_{term}.

Parameter | Result with regimen^{a}: | P (ANOVA) | |||||||
---|---|---|---|---|---|---|---|---|---|

Quinine for 7 days (Q) | Artemisinin + 3 days of quinine (AQ3) | Artemisinin + 5 days of quinine (AQ5) | |||||||

Radical cure (n = 56) | Recrudescence | Radical cure (n = 44) | Recrudescence | Radical cure (n = 66) | Recrudescence (late; n = 11) | ||||

Late (n = 8) | Early (n = 3) | Late (n = 17) | Early (n = 11) | ||||||

Observed values | |||||||||

Geometric mean P(0) (per μl) | 16,282 (11,722–22,617) | 20,317 (8,646–47,740) | 13,111 (5,651–30,420) | 17,541 (12,123–25,379) | 26,733 (14,344–49,823) | 13,461 (6,645–27,267) | 24,917 (18,619–33,345) | 16,987 (5,520–52,279) | 0.429 |

Mean PCT (h) | 58 (53–64) | 72 (51–93) | 107 (76–137) | 40 (35–45) | 46 (40–52) | 52 (41–63) | 42 (41–46) | 48 (34–62) | 0.004 |

Mean ET (h) | 110 (104–115) | 96 (75–117) | 61 (31–92) | 39 (34–44) | 32 (26–38) | 26 (15–37) | 84 (80–88) | 75 (60–89) | 0.000 |

Model-estimated values | |||||||||

Mean lag time (h) | 5 (3–8) | 5 (0–12) | 14 (12–16) | 3 (2–5) | 2 (1–4) | 5 (1–10) | 3 (2–4) | 7 (2–13) | 0.003 |

Mean t_{1/2el}(h) | 7.8 (7.3–8.3) | 8.6 (7.4–9.8) | 11.4 (6.7–16.2) | 4.7 (4.5–4.9) | 5.0 (4.7–5.4) | 5.5 (4.8–6.2) | 5.3 (5.1–5.5) | 5.8 (5.0–6.5) | 0.000 |

Mean PCT_{calc}(h) | 92 (85–99) | 102 (86–119) | 142 (76–207) | 55 (51–58) | 60 (55–65) | 66 (56–77) | 63 (59–66) | 69 (58–81) | 0.000 |

Geometric mean P_{term} (per μl) | 0.003 (0.001–0.008) | 0.021 (0.003–0.156) | 0.969 (0.005–183) | 0.138 (0.074–0.255) | 0.396 (0.160–0.982) | 0.959 (0.238–3.862) | 0.001 (0.000–0.002) | 0.003 (0.000–0.038) | 0.000 |

Mean replication t_{1/2}(h) | 21 (13–29) | 22 (−20–63) | 39 (34–44) | 32 (14–50) | 23 (18–28) | 0.029 |

^{a}

*P*

_{term}.

*P*

_{term}. The same was the case for

*P*(0) or the estimated intercept and the estimates of

*k*(and thus

*t*

_{1/2el}) and

*P*

_{term}. These were all entered separately into the Cox model.

*P*

_{term}was shown to be a significant predictor of recrudescence (P < 0.001; relative risk for 1 log increase of

*P*

_{term}, 1.7 [95% CI, 1.4 to 2.1]), but PCT and ET were also associated with a greater hazard function of recrudescence (data not shown).

*P*(0), PCT, ET, and

*P*

_{term}were entered separately in the model. No more than 80% of cases were classified correctly as recrudescence or radical cure, with no important differences between the respective parameters.

*P*

_{term}is lower for the patients with radical cure than for the patients with recrudescence. The model did not discriminate between early and late recrudescence. In group AQ5, there was no difference between radical cure and late recrudescence with respect to

*P*

_{term}.

*P*

_{term}evolves into recrudescence was 0.03 h

^{−1}on a natural logarithmic scale, which corresponds to a 0.013-log increase per h, or a 0.6-log increase per 48-h cycle. There was a slight difference in replication half-life between the subgroups of recrudescence.

## DISCUSSION

*P*

_{term}is an important determinant of the chance of recrudescence.

*P*

_{term}: the estimate of the initial parasitemia,

*k*, and the duration of therapy.

*P*(0) belonged to the inclusion criteria, and therefore it was within a relatively narrow range. The decline of the parasitemia started earlier after the artemisinin regimens than after quinine. This illustrates the fast antiparasitic activity of artemisinin and the great range of the parasite development cycle on which it exerts its action, which confirms in vitro findings (14). The impact of the lag phase on

*P*

_{term}is small though. The parasite elimination rate is more important, and when this is slow, the eradication time, and thus duration of therapy, becomes critical. A slower parasite clearance, and thus a higher value of

*P*

_{term}, was associated with recrudescence in groups Q and AQ3. It was also shown that the longer eradication time in regimen AQ5 lowers

*P*

_{term}and improves efficacy, in comparison to those of regimen AQ3. In regimen AQ5 itself, there was no difference between radical cure and recrudescence with respect to

*P*

_{term}. The lack of this difference is not clear, but it should be noted that a late recrudescence could not be discriminated from a reinfection. In the low-transmission study area, reinfection is probably infrequent, and an extra argument is that the replication rate in the cases of recrudescence attained a realistic value. A more plausible explanation for the lack of difference in

*P*

_{term}between radical cure and recrudescence in regimen AQ5 is at the same time the Achilles heel of the kinetic model. During the first 24 to 48 h of therapy, the elimination rate constant is dominated by the effects of artemisinin. In this period, the exponential decline is an adequate description of the time course of the parasitemia. However, later, when the clearance rate slows down to that of quinine, a second elimination constant should be incorporated into the model. The effect of a second elimination constant on the parasitemia was not detected by the model, probably because by that time the parasitemia has decreased to or is under the detection level in most cases. However, there was a significant difference in elimination rates after artemisinin or quinine. The low precision of the low parasite counts and the relatively high limit of detection are important limitations for more refined kinetic modeling.

*P*

_{term}. This does not provide a satisfactory algorithm with which to predict recrudescence or to individualize the duration of therapy. With our current knowledge, the best advice for an individual patient is to aim at an eradication time of at least 3 days with quinine after a single dose of artemisinin. A design in which patients would be randomized to eradication time could give a better idea of the possibilities of individualization of treatment.

*Plasmodium falciparum*are not yet fully understood. However, White recently presented arguments, based on pharmacodynamic concepts, explaining how mefloquine resistance developed so quickly in Thailand (15).

*P*

_{term}is an important determinant for outcome: radical cure or recrudescence. It suggests that there is a point of no return the actual value of which may depend on treatment regimen.

*P*

_{term}could only be studied because of the short residence time of artemisinin and, to a lesser extent, of quinine. With chloroquine and mefloquine, the extremely long residence times preclude accurate estimation of the eradication time. Nevertheless, we feel that these concepts of parasite kinetics can be generalized to other drugs also and that they may provide tools for a rational design of new antimalarial treatment regimens.

## ACKNOWLEDGMENT

## Appendix

### Population kinetics of parasitemia.

*P*(

*t*), excluding the negative blood smears. It appeared that for most individuals the decline of ln

*P*(

*t*) was more or less constant over all 8-h intervals, with the exception of the first interval. In a plot of the geometric mean values (Fig. FA1), this is less clear, because the tail of the mean curve is distorted by blood smears becoming negative. So it seemed rational to start with a simple log-linear (exponential) decline of the parasitemia as the basic model and build from this. In the formula

*A*is the estimated initial parasitemia and

*k*is the elimination rate constant.

*k*can be recalculated into a more conventional elimination half-life,

*t*

_{1/2el}, according to the formula

*k*was allowed to change per treatment regimen, was also constructed, and this model gave the best fit (BIC = 4,720.6). Negative blood smears, with a value of zero, were not included in the models. The three models were also fitted to ln [

*P*(

*t*) + 0.5], with zero values included, but this did not have a significant effect on the BIC and estimates. In further modeling, zero values were excluded from the data set.

*A*, was allowed to change per regimen. As expected (treatment groups were similar with respect to baseline parasite count), this did not improve the fit, so that the variation of this term could be interpreted as a random effect.

*A*was greater than the observed initial parasitemia,

*P*(0), or put simply, after drug intake, it takes some time before the parasitemia starts to decline. This time can conceptually be simplified to a lag phase,

*t*

_{lag}. Although the lag phase is not readily explained in biologically plausible terms, in clinical experience, a lag phase is usually interpreted as the time until

*P*(

*t*) has decreased to values lower than

*P*(0). Moreover, the definitions of the in vivo response to drug treatment are based on decrease of the parasite count relative to the initial parasitemia, thereby ignoring that, in many patients, the parasitemia increases initially and that the lag times may be different for individuals. When the lag time is incorporated into the kinetic models, these have the form

*A*and

*t*

_{lag}are interdependent, which means that a difference in

*t*

_{lag}also affects the value of

*A*. Nevertheless, the models were fitted, again incorporating

*t*

_{lag}, and this did not yield better fits than model I.

*c*and

*k*changing per regimen, or with only

*k*changing per regimen. The latter two models were also investigated with

*k*as a fixed factor. Model II, the model with a fixed quadratic term,

*c*, for every patient and a linear term,

*k*, variable per patient, but depending on regimen, yielded the best fit (BIC = 4,719.4). In both models I and II, the estimates of

*k*were comparable for regimens AQ3 and AQ5 and different from those for regimen Q. The difference in BIC for models I and II was small, and model II had 1 df more than model I. Another approach to describe the initial rise in the parasite count was to build models in which the logarithm of

*t*was factored in. The basic model of this approach looks like

*t*is described by

*s*fixed for all patients, independent of treatment regimen, and

*k*, with random variation, but with a value depending on the regimen, yielded the best fits (BIC = 4,640) of this group of models.

*s*is the stage in the parasites' cycle.

*c*and

*s*, respectively, it was decided that the simplest model, model I, would be taken as the best description for the data. In this model, the concept of

*P*

_{term}, fitted by this model, gave a plausible explanation of the mechanism of recrudescence.

## REFERENCES

*Trans. R. Soc. Trop. Med. Hyg.*84 1990 499 -502

*Pharm. Pharmacol. Lett.*3 1996 127 -130

*Am. J. Trop. Med. Hyg.*55 1996 438 -443

*WHO Bull.*76 (Suppl.1) 1998 51 -58

*Drugs*52 1996 818 -836

*Trans. R. Soc. Trop. Med. Hyg.*90 1996 61 -65

*Trans. R. Soc. Trop. Med. Hyg.*85 1991 210 -211

*Trans. R. Soc. Trop. Med. Hyg.*91 1997 191 -194

*Lancet*337 1991 1140 -1143

*Trans. R. Soc. Trop. Med. Hyg.*88 1994 324 -327

*Recent advances in tropical medicine*1928 79 -80 J.&-A. Churchill London, United Kingdom

*Ann. Stat.*6 1978 461 -464

*Am. J. Trop. Med. Hyg.*48 1993 398 -402

*Exp. Parasitol.*76 1993 85 -95

*Antimicrob. Agents Chemother.*41 1997 1413 -1422

*Phil. Trans. R. Soc. Lond. B*354 1999 739 -749

*Trans. R. Soc. Trop. Med. Hyg.*86 1992 590 -597

*Am. J. Trop. Med. Hyg.*47 1992 112 -116

## Information & Contributors

### Information

#### Published In

#### Copyright

#### History

### Contributors

## Metrics & Citations

### Metrics

**Note: **There is a 3- to 4-day delay in article usage, so article usage will not appear immediately after publication.

Citation counts come from the Crossref Cited by service.

### Citations

If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.