Logistical feasibility and potential benefits of a population-wide passive-immunotherapy program during an influenza pandemic

Abstract

Treatment strategies for severe cases of pandemic influenza have focused on antiviral therapies. In contrast, passive immunotherapy with convalescent blood products has received limited attention. We consider the hypothesis that a passive-immunotherapy program that collects plasma from a small percentage of recovered adults can harvest sufficient convalescent plasma to treat a substantial percentage of severe cases during a pandemic. We use a mathematical model to estimate the demand and supply of passive immunotherapy during an influenza pandemic in Hong Kong. If >5% of 20- to 55-year-old individuals recovered from symptomatic infection donate their plasma (donor percentage > 5%), >67% of severe cases can be offered convalescent plasma transfusion (treatment coverage > 67%) in a moderately severe epidemic (R ₀ < 1.4 with 0.5% of symptomatic cases becoming severe). A donor percentage of 5% is comparable to the average blood donation rate of 38.1 donations per 1,000 people in developed countries. Increasing the donor percentage above 15% does not significantly boost the convalescent plasma supply because supply is constrained by plasmapheresis capacity during most stages of the epidemic. The demand–supply balance depends on the natural history and transmission dynamics of the disease via the epidemic growth rate only. Compared to other major cities, Hong Kong has a low plasmapheresis capacity. Therefore, the proposed passive-immunotherapy program is a logistically feasible mitigation option for many developed countries. As such, passive immunotherapy deserves more consideration by clinical researchers regarding its safety and efficacy as a treatment for severe cases of pandemic influenza.

Influenza pandemic plans for treating severe cases that require hospitalization or intensive care currently focus on the use of antivirals, anti-inflammatory agents, and nonpharmacologic means, such as optimal ventilator and fluid management. In contrast, passive immunotherapy has received limited attention (1). A recent meta-analysis study suggested that during the 1918 influenza pandemic, transfusion of blood products from recovered individuals reduced the mortality rate of severe cases by more than 50% (2). The proof of principle for this therapeutic approach was demonstrated in the modern clinical setting when a patient in Shenzhen, critically ill with avian influenza A/H5N1 virus infection in 2006, was administered convalescent plasma from a woman who had survived infection (3). The patient, whose condition was worsening despite treatment with oseltamivir, recovered after receiving several infusions of convalescent plasma. In 2007, Simmons et al. (4) used a murine model of infection to demonstrate the prophylactic and therapeutic efficacy of monoclonal antibodies with neutralizing activity produced by immortalized B lymphocytes of convalescent patients recovering from influenza A/H5N1 virus infection in Vietnam, thereby providing further preclinical evidence that passive immunotherapy with convalescent blood products could be a practical treatment option during an influenza pandemic. Unlike antivirals, the use of neutralizing antibodies would not be affected by mutations associated with viral resistance to neuraminidase inhibitors or adamantanes. Although antigenic drift may gradually lower its effectiveness, significant antigenic drift nullifying its effectiveness is unlikely to occur within the timescale of a pandemic wave.

A Population-Wide Passive Immunotherapy Program.

In this article, we assume that convalescent plasma (CP) is efficacious in treating severe cases of pandemic influenza. Under this premise, we test the hypothesis that a population-wide passive-immunotherapy program that collects plasma from a small percentage of convalescent individuals can harvest sufficient CP to treat a substantial percentage of severe cases during the first wave of the pandemic. The proposed program involves recruiting adults (individuals aged 20–55 years who represent a proportion n_D of the population) to donate blood if they have experienced influenza-like symptoms more than 2 weeks prior (to account for the time needed for neutralizing antibodies to build up). The blood samples would be screened for infectious diseases (including HIV, hepatitus B and C viruses, human T-lymphocite virus, syphilis, and so forth, as in routine blood donation screening) and neutralizing antibodies against the pandemic virus. Donors whose blood samples are free of known infectious agents and contain a sufficiently high titer of neutralizing antibodies would then be invited to donate plasma by plasmapheresis or routine whole blood donation. Qualified donors with higher titers may be given higher priority for plasma donation.

CP can be used in two types of passive immunotherapy: (i) direct transfusion of CP and (ii) administration of hyperimmune i.v. immunoglobulins (IVIG) extracted from CP fractionation. The two options have their strengths and weaknesses. The major advantage of CP transfusion in the form of fresh frozen plasma is that it is an inexpensive and common procedure with a lead time (the time between plasma donation and availability of quality-checked CP for treatment) of no more than a couple of weeks. However, the neutralizing antibodies concentration in CP cannot be easily adjusted. Moreover, the treatment dosage is limited by the fluid load that CP transfusion imposes on the recipient. In contrast, hyperimmune IVIG is much more controllable in dosing. Furthermore, hyperimmune IVIG has a lower fluid volume and a lower risk of transfusion-transmitted infection, allergic reactions, and allo-immunization after additional steps of treatment to remove irrelevant plasma protein and to inactivate possible pathogens during plasma fractionation. Such hyperimmune IVIG preparation is likely to be safer for a critically ill patient with lungs flooded by inflammatory exudates with precarious fluid balance. A major drawback of hyperimmune IVIG, however, is that its preparation relies on fractionation technology, which is more costly and requires longer lead time. Moreover, sufficient CP must be collected before the fractionation process can be started.

In this article, we use the demographic and logistical parameters of Hong Kong (a Southeast-Asian metropolitan area with a population of 7 million) as a case study. See Fig.1 for a schematic of the proposed passive-immunotherapy program. We examine the following questions regarding the logistical feasibility and potential benefits of the proposed passive immunotherapy program: (i) What percentage of convalescent individuals (donor percentage) is needed in order for the program to significantly reduce pandemic mortality? (ii) How many severe cases can be offered passive immunotherapy? (iii) What are the rate-limiting factors in the supply of passive immunotherapy? (iv) What are the epidemiologic and logistical factors that determine the demand-supply balance of passive immunotherapy?

Fig. 1.

Schematic of the proposed passive immunotherapy program.

Transmission and Natural History Model for Pandemic Influenza.

We use an age-structured disease transmission model to simulate the spread of pandemic influenza. The natural history model is similar to that used by Longini et al. (5, 6). See Methods and SI Methods for details. The most important parameter in characterizing the growth of an epidemic is the basic reproductive number R ₀, which is defined as the average number of secondary cases generated by a typically infectious individual in a completely susceptible population. We consider values of R ₀ between 1.2 and 2, which is consistent with recent estimates (5, 7–9).

Logistical Model for the Passive-Immunotherapy Program.

We assume that q_D (%) of 20- to 55-year-old individuals, who have recovered from symptomatic infections of pandemic influenza, donate their blood for screening T_R = 14 days after cessation of symptoms. Follow-up of convalescent individuals infected with H1N1pdm in an ongoing clinical trial of passive immunotherapy (of which K.Y.Y. is the principal investigator and C.K.L. a coinvestigator) suggested that neutralizing antibody levels reach maximal level around 14 to 21 days after recovery and stays at that level for months after. We assume that q_S (%) of these donors are qualified for plasma donation, of which q_R (%) are recurrent donors who return to donate plasma every T_W = 14 days. Screening involves both detection of infectious agents and neutralizing antibodies against the pandemic virus. The latter is the rate-limiting step because neutralization tests of pandemic viruses can only be done in a biosafety level 3 setting. We assume that five biosafety level 3–trained technicians are available to test the blood specimens, each running 150 viral neutralization tests in 3 days. Therefore, the capacity and turnaround time of blood screening are U_S = 750 and T_S = 3 days, respectively. Hong Kong currently has nine plasmapheresis machines, which allow a maximal throughput of 162 plasma donations per day (assuming 12-h daily operation, with each donation taking 40 min). Therefore, the capacity and turnaround time of plasmapheresis are U_P = 9 and T_P = 1/18 days, respectively.

Collected CP are ready for use in transfusion after final quality check, which takes T_Q = 2 days. Preparation of hyperimmune IVIG takes T_IVIG more days from that point on. We assume that r_T plasma donations are required to treat one severe case on average. The expert panel of the above-mentioned ongoing clinical trial of passive immunotherapy for H1N1pdm in Hong Kong suggested that r_T < 10 and T_IVIG ranges from 30 to 90 days.

Demand for Passive Immunotherapy.

We assume that p_H (%) of symptomatic cases will be severe cases for whom passive immunotherapy is suitable. Although p_H will be smaller than the case-hospitalization rate (passive immunotherapy may not be suitable for some hospitalized cases), we assume that the two have similar ranges and consider p_H ranging from 0.1 to 1%. Because each severe case requires r_T plasma donations on average, demand for CP is simply r_Tp_H times the number of symptomatic cases. Therefore, r_Tp_H can be regarded as a single parameter, which we refer to as the “lumped-demand parameter.”

Outcome Measure.

We define the outcome as the percentage of severe cases that can be offered passive immunotherapy by the proposed program during the first wave of the local epidemic. We refer to this outcome as “treatment coverage” and denote it by ρ.

Results

When presenting our results, we will focus on the case where CP transfusion is the choice of passive immunotherapy. The system dynamics for the case of hyperimmune IVIG is identical, but with a longer lead-time.

General Dynamics.

We first show the general pattern of daily demand and supply of CP treatment over the course of the epidemic (Fig. 2). During the early phase, daily demand and supply both increase exponentially at the initial epidemic growth rate r: this is because both are driven by the flow of symptomatic cases, and both screening and plasmapheresis have not yet reached their maximal capacity (i.e., availability of donors is the bottleneck of CP supply). During this phase, recurrent donors have little impact on supply because the number of first-time donors for plasmapheresis is increasing exponentially, and the resting period of 14 days required for recurrent donations is long at the epidemic time-scale. Daily supply exceeds daily demand if

in which case a stockpile of CP is built from the surplus (SI Methods). The exponential phase ends when either plasmapheresis (case 1) or screening (case 2) reaches maximal capacity. Cases 1 and 2 correspond to the conditions q_SU_S/T_S > U_P/T_P and q_SU_S/T_S < U_P/T_P [i.e., whether q_S is larger or smaller than U_PT_S /U_ST_P = 65% in our base case (Table 1)]. These conditions imply that the exponential phase ends earlier in case 1, for which daily supply plateaus at the maximal plasmapheresis throughput U_P/T_P immediately after. In case 2, after the exponential phase, daily supply increases linearly at rate q_Rq_SU_S/T_ST_Wr_T until reaching its plateau at U_P/T_P (SI Methods). The linear phase is driven by the buildup of recurrent donors when screening has reached its maximal capacity but plasmapheresis has not. If there are no recurrent donors (q_R = 0), then the linear phase disappears and the daily supply plateaus at the maximal throughput of qualified donors from screening (i.e., q_SU_S/T_S). These results show that recurrent donors are an important source of CP supply in case 2 (when q_S is small) but not necessarily in case 1 (when q_S is large).

Fig. 2.

General demand-supply dynamics of passive immunotherapy under the proposed program. Key parameter values are basic reproductive number R ₀ = 1.4, donor percentage q_D = 15% and lumped-demand parameter r_Tp_H = 1.5%. (A) Daily demand and supply when the percentage of donors qualified for plasmapheresis is q_S = 80% (case 1) and q_S = 30% (case 2). The same colors for the two cases are used in B and C. (B) Daily number of severe cases treated (solid lines) and stockpile of CP built from supply surplus (dashed line). There is no CP stockpile in case 2. (C) The percentage of daily demand met. Over the course of the epidemic, 82% and 58% of total demand are met in case 1 and 2, respectively.

Table 1.

Model parameters

Parameter	Description	Base case value	Sensitivity range	Sources
R0	Basic reproductive number	1.2–2	1.2–2	(6, 9,16-17)
DE	Mean latent duration	1.2 days	0.5–2 days	(5, 6, 18)
DA	Mean asymptomatic duration	4.1 days	1–5 days	(5, 6, 18)
DI	Mean symptomatic duration	4.1 days	1–5 days	(5, 6, 18)
hA	Relative infectiousness of asymptomatics	0.5	0–2	(5, 6, 18)
pS	Percentage of infected cases who are symptomatic	67%	50–100%	(5, 6, 18)
pH	Percentage of symptomatic cases hospitalized	0.1–1%	0.1–2%	(19, 20)
qD	Donor percentage: The percentage of 20- to 55-year-old convalescent individuals who are willing to donate plasma for the passive immunotherapy program	5–50%	5–50%	Assumed*
qS	Percentage of screened donors whose plasma are qualified for use in passive immunotherapy	80%	10–100%	Assumed*
qR	Percentage of qualified donors who are recurrent donors	20%	0–50%	Assumed*
TR	Minimal waiting time (for buildup of antibodies) before convalescent individuals are ready for screening	14 days	14 days	Assumed*
TW	Minimal waiting time after plasmapheresis before recurrent donors can undergo plasmapheresis again	14 days	14 days	Assumed*
US	Number of screening servers for testing blood samples	150 × 5 = 750	750	Assumed*
TS	Average turnaround time for blood sample screening	3 days	3 days	Assumed*
UP	Number of plasmapheresis machines	9	9	Assumed*
TP	Average time for plasmapheresis (per donation)	40 min	40 min	Assumed*
TQ	Average time for final quality check of CP	2 days	2 days	Assumed*
rT	Average number of donations needed per case treated	2–10	2–10	Assumed*

^*Assumptions are based on our experience in an ongoing trial of passive immunotherapy for pandemic influenza, of which K.Y.Y. is the principal investigator and C.K.L. is a coinvestigator.

Base Case.

Next, we consider the base case scenarios in Table 1, assuming q_S = 80% (i.e., q_S > U_PT_S/U_ST_P = 65%; case 1). In general, the treatment coverage ρ increases sharply as the basic reproductive number R ₀ and the lumped demand parameter r_Tp_H decrease (Fig. 3A). In particular, when R ₀ is large and r_Tp_H is small, ρ is very sensitive to r_Tp_H but insensitive to R ₀. Similarly, when R ₀ and r_Tp_H are small, ρ is very sensitive to both. With a donor percentage of q_D = 15%, the proposed program can supply passive immunotherapy to more than 82% of severe cases (ρ > 82%) if R ₀ < 1.4 and r_Tp_H < 1.5% but less than 35% if R ₀ > 1.8 and r_Tp_H > 1.5%. In general, the treatment coverage ρ increases sharply as the donor percentage q_D rises from 0% but with rapidly decreasing marginal increase (Fig. 3B). When R ₀ < 1.4 and r_Tp_H < 1.5%, ρ > 67%, even if q_D is as low as 5%, which is comparable to the current average blood donation rate of 38.1 donations per 1,000 people in developed countries (10). When q_D is greater than 15%, ρ becomes largely insensitive to further increase in q_D in most scenarios. The treatment coverage ρ for q_D = 15% is more than 81% of that for q_D = 50% across all values of R ₀ and r_Tp_H considered in the base case. Therefore, increasing the donor percentage q_D beyond 15% has a relatively small impact on CP supply. This is because increasing q_D can boost supply only during the exponential phase (Fig. 2A) when plasmapheresis is not yet the supply bottleneck. For the same reason, once the donor percentage q_D has reached 15%, the treatment coverage ρ is insensitive to further increase in q_D, even when the plasmapheresis and screening capacity are doubled (Fig. 3B, Lower).

Fig. 3.

Treatment coverage in the base case. (A) Treatment coverage ρ when the donor percentage is q_D = 15%. (B) Sensitivity of ρ against q_D with R ₀ = 1.4 in the base case (Upper) and when the plasmapheresis and screening capacity are doubled (Lower). In general, increasing q_D beyond 15% has little impact on the outcome.

Sensitivity Analysis.

We conduct an extensive multivariate sensitivity analysis to test the robustness of our base-case observations against uncertainties in parameter values. We generate 15,000 epidemic scenarios by randomly selecting parameter values from their plausible ranges using Latin-hypercube sampling (Table 1). Although there are numerous model parameters, the treatment coverage ρ is mainly determined by three lumped parameters (Fig. 4): (i) r_Tp_H, which indicates the magnitude of demand; (ii) q_Sq_D, which indicates the magnitude of supply; and (iii) the initial growth rate of the epidemic r. Although the dependence of ρ on r_Tp_H and q_Sq_D is readily comprehensible, it is not obvious a priori that ρ depends on the natural history and transmission dynamics of the disease via only the initial epidemic growth rate. When the plasmapheresis and screening capacity are very large, the supply-demand dynamics are further simplified: the treatment coverage ρ depends on the lumped demand parameter r_Tp_H and the lumped supply parameter q_Sq_D only via their ratio (SI Methods, Fig S1). Finally, Fig. 4 shows that ρ becomes insensitive to q_Sq_D when the latter increases beyond 15 to 20%, thereby generalizing our base case observations in Fig. 3B.

Fig. 4.

Treatment coverage for 15,000 epidemic scenarios generated using Latin-hypercube sampling with different level of demand parameters (A–D). Initial epidemic growth rate is expressed as initial epidemic doubling time on the x axis for ease of interpretation. Each point corresponds to a randomly generated epidemic scenario and is colored with the associated treatment coverage. The relatively small amount of overlapping of different colors indicate that the treatment coverage is mostly determined by the abscissa (the doubling time) and ordinate (the supply parameter q_Sq_D).

Hyperimmune IVIG.

The treatment coverage ρ deteriorates rapidly as the lead-time for hyperimmune IVIG (T_IVIG) increases (Fig. 5). Having a lead-time of T_IVIG is equivalent to shifting the supply curve in Fig. 2A to the right by T_IVIG. Consequently, the drop in treatment coverage ρ becomes steeper as the initial growth rate of the epidemic increases. The larger the left-hand side of Eq. (1), the smaller the drop in the treatment coverage ρ because of T_IVIG. In general, however, with a lead-time of at least 50 days, the treatment coverage is at least halved when the doubling time of the epidemic is shorter than 9 days and the lumped-demand parameter r_Tp_H > 0.9%. For example, when the initial doubling time is 10 days and the donor percentage is q_D = 15%, the treatment coverage with r_Tp_H = 0.9% and T_IVIG = 50 days is reduced from 82 to 39% when compared to the case of CP transfusion.

Fig. 5.

The impact of hyperimmune IVIG lead-time on treatment coverage. In general, treatment coverage decreases rapidly as T_IVIG increases unless the epidemic growth rate is very slow.

Discussion

Our results suggest that with plasmapheresis capacity similar to that in Hong Kong, the proposed passive-immunotherapy program can supply CP transfusion to treat 67 to 82% of severe cases in a moderate pandemic (basic reproductive number R ₀ < 1.4, lumped-demand parameter r_Tp_H < 1.5%) when the donor percentage is 5 to 15%. Increasing the donor percentage beyond 15% has little additional benefit because CP supply is constrained by the capacity of plasmapheresis during most stages of the epidemic. Increasing plasmapheresis capacity could significantly boost CP supply, especially when there is a substantial pool of recurrent donors to alleviate the dependence of CP supply on donor percentage. In an ongoing clinical trial of passive immunotherapy for H1N1pdm virus infection in Hong Kong, 20% of convalescent individuals agreed to donate their plasma for the study. Therefore, the donor percentage required by the proposed passive-immunotherapy program (5–15%) is likely to be feasible. In view of the logistical feasibility of such a program, we recommend that further clinical studies are conducted to evaluate the safety and efficacy of passive immunotherapy as a treatment for severe cases of pandemic influenza virus infection.

Our study is based on the premise that CP will be efficacious in reducing morbidity and mortality associated with pandemic influenza. In theory, the polyclonal nature of neutralizing antibodies in CP would lower the probability of an escape mutant emerging in treated patients. Furthermore, besides providing neutralizing antibodies against the pandemic virus, CP might also carry antibodies to other bacterial pathogens, which might decrease the severity of coexisting bacterial infections (4). As such, CP not only might reduce the case fatality rate but might also increase the recovery rate and shorten duration of hospitalization of severe cases. The proposed passive-immunotherapy program can thus significantly reduce the burden on the healthcare system, especially the intensive care unit, which will likely be stressed, if not overloaded, at the peak of an influenza pandemic wave, hence benefiting the general public and not only those receiving passive immunotherapy. Although the hypothesized efficacy of CP has yet to be proven in clinical trials, our modeling results show that a public health system similar to that in Hong Kong has the capacity to support a population-wide passive-immunotherapy program that can supply CP treatment to a substantial percentage of the severe cases in a moderately severe pandemic. We estimate that compared to other developed countries, Hong Kong has a relatively low plasmapheresis capacity. Our conclusions regarding donor percentage needed and rate-limiting factors remain valid for plasmapheresis capacity ranging from 50 to 400% of what we have assumed in the base case. See SI Methods, Figs. S2–S5, and Table S1 for details.

Our conclusions are robust against uncertainties in the natural history and transmission dynamics of pandemic influenza. Our sensitivity analysis shows that the outcome depends on these epidemiological characteristics only via the initial growth rate of the epidemic r (Fig. 4). As such, our results are applicable not only to pandemic influenza but also to other emerging infectious diseases for which the time-scales of disease transmission and antibody response are similar to that for influenza virus. The three determinants of treatment coverage (the initial epidemic growth rate r, the lumped demand parameter r_Tp_H, and the lumped supply parameter q_Dq_S) are all readily measurable in real time during an epidemic. Therefore, charts similar to those in Fig. 4 can be used as a general reference for estimating the treatment coverage of the proposed passive immunotherapy program for a given plasmapheresis capacity.

Although we did not explicitly model the effect of pandemic mitigation strategies, our results are likely to be insensitive to the incorporation of such because the transmission-reducing effect of control measures is approximately equivalent to a decrease in epidemic growth rate. We have assumed that transmissibility of the virus is constant throughout the epidemic. If transmissibility becomes lower at some epidemic stages, either because of deliberate public health interventions or seasonal forcing or other factors, the treatment coverage will be higher than predicted here. That result is because demand would be lower because of slower transmission, although supply would be largely unaffected because it is constrained during most of the epidemic by plasmapheresis capacity but not the flow of new donors, especially when there is a substantial pool of recurrent donors. The converse holds in the case of higher transmissibility.

In theory, antiviral therapy might lower the level of neutralizing antibodies in convalescent plasma because of its effectiveness in reducing the duration of viral shedding and illness. However, in a randomized controlled trial of antiviral treatment of individuals with seasonal influenza virus infection, the placebo group and two groups receiving antiviral treatment all experienced on average 16-fold rises in antibody titers between baseline and convalescence (11). If antiviral therapy does lower the level of neutralizing antibodies in CP, then the percentage of qualified donors q_S would be smaller, and the associated impact on the proposed passive immunotherapy program can be estimated using Fig. 4. To preserve the effectiveness of the program, effort should be made to retain qualified donors as recurrent donors.

If hyperimmune IVIG is preferred to CP transfusion as the standard passive immunotherapy for the program, the treatment coverage is substantially reduced because the associated lead-time is likely to be at least 2 months. In this case, the population-wide passive-immunotherapy program will have limited benefit, regardless of plasmapheresis capacity and donor percentage, unless the initial doubling time of the epidemic is much longer than 10 days or the proportion of severe cases is very small.

In our model, we have only considered the first wave of a pandemic for a closed population in which disease transmission has no spatial heterogeneity. In large countries, such as the United States, epidemics will not be completely synchronized across different geographical regions. In this context, a region-wide cooperative program may be considered in which upstream populations (populations that experience an epidemic first) could supply CP to downstream populations, perhaps in exchange for other medical resources (e.g., antivirals, masks, and so forth) from the latter in return. In the case of multiple pandemic waves (e.g., because of seasonal forcing), there will be more time for preparation of CP and donors infected in the first wave could become an important source of CP for use in subsequent waves.

Methods

Both the disease transmission model and the queuing model are formulated as differential equation systems. See SI Methods for the model equations. The population is stratified into age groups of 5 years (0–4, 5–9, 10–14, 15–19, 20–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69, 70+) . The who-acquires-infection-from-whom matrix is constructed using social contact data (12). Similar transmission models have been used in recent modeling studies of influenza (13, 14). The waiting times T_W, T_R, T_S, and T_IVIG in our model are likely to have small variance. As such, to relax the exponential waiting time distribution assumption imposed on these durations by the differential equation formulation, we divide these waiting times into substages with equal mean durations so that the overall waiting times are gamma-distributed with small variance (15).