Design and optimization of an HVAC-integrated ozone disinfection system for the purified buffer room of PIVAS Authors: Chun-Yang Zhang, Xiang-Lian Li, Dan-Yan Shao, Jin-Qiu You, Jian-An Bao, Jing-Jing Ma Published: 2026 ABSTRACT SummaryThe limitations of traditional disinfection protocols in pharmacy intravenous admixture services (PIVASs) render it necessary to develop advanced disinfection strategies to guarantee medication safety. This study introduces an ozone disinfection system incorporated into the upstream section of the heating, ventilation, and air conditioning (HVAC) system for the purified buffer room of PIVASs. By conducting a comparative analysis of airborne particles, viable airborne bacteria, settled bacteria, and surface microbial loads between conventional areas and the ozone-disinfected buffer room, in conjunction with response surface methodology (RSM) modeling, the results indicate significant benefits of the ozone-integrated buffer room design for PIVAS applications. The optimal disinfection parameters (an ozone concentration of 8 ppm with a 67-minute exposure) effectively enhance the environmental quality of PIVASs, reduce the risks of medication contamination, and offer strong support for the safe operation of PIVAS facilities. Introduction The preparation of clinical medications in non-sterile settings presents significant risks, encompassing drug contamination, incompatible drug combinations, adverse drug reactions (ADRs), cross-infection, and cross-resistance.1,2 Pharmacy intravenous admixture services (PIVASs), a centralized facility designed for the preparation and distribution of intravenous (IV) solutions through strict aseptic techniques within controlled sterile environments, was developed to meet the evolving demands of healthcare.3,4,5 The first PIVAS was established at the Ohio State University Hospital in 1969, while China’s first PIVAS was implemented at Shanghai Jing’an District Central Hospital in 1999. With the rapid development and expansion of PIVAS in China, strict standards for PIVAS construction and management have been put into effect, with “hygiene and disinfection” emerging as a core requirement.6 Buffer design was extensively implemented in infection control facilities during the COVID-19 pandemic, effectively reducing the risks of cross-infection. In light of this evidence, we put forward a design of a purified buffer room equipped with mobile air purifiers. Unpacked medications are transferred to individual compounding rooms through pass-through windows, with the exception of dedicated antibiotic, nutrition, and chemotherapy compounding suites. Moreover, ozone generators are incorporated into the upstream section of the HVAC system for these rooms, facilitating comprehensive disinfection of functional areas. Ozone disinfection is widely employed in medical environments, food preservation, and water purification due to its strong oxidizing properties,7,8,9 which exert broad-spectrum inactivation effects on viruses, bacteria, fungi, protozoa, and other microorganisms. Its microbicidal mechanism primarily entails oxidative reactions with key cellular components (e.g., proteins, lipids, and nucleic acids), which disrupt the structure and function of microorganisms.10 As an environmentally friendly disinfectant, ozone is transformed into molecular oxygen during the oxidation-disinfection process.11 Leveraging the ozone-integrated disinfection design for A purified buffer room, this study evaluates the efficacy of the purified buffer system in PIVAS through comparative analysis of dynamic environmental monitoring data between the buffer room and public areas. It further optimizes ozone against common bacterial colonies on typical PIVAS materials.12,13,14 The results demonstrated that the buffer room significantly improved the cleanliness level of unpackaged medications. Using response surface methodology (RSM) and LINGO parameter optimization, the optimal ozone disinfection conditions (8 ppm concentration for 67 min) were identified, which were subsequently validated in a small test chamber to exhibit excellent bactericidal efficacy against Staphylococcus aureus and Escherichia coli. This design establishes a PIVAS cleanliness management framework (“PIVAS environment → purified buffer room → compounding rooms”), providing robust cleanliness assurance for PIVAS infrastructure development. Results Evaluation of the air disinfection effect of ozone disinfection in the added purification buffer room in PIVAS The critical area of our hospital’s PIVAS center consists of clean and non-clean control zones, including specialized functional rooms: antibiotic compounding room, nutrition compounding room, chemotherapy compounding room, purification buffer room, and public area (Figure 1A). To verify the effectiveness of the design integrating ozone disinfection in the purification buffer room, we quantified airborne bacteria, settled bacteria, and airborne particles in both the buffer room and the public area after daily cleaning and disinfection. The average airborne bacterial count was 6 CFU in the buffer room, compared to 28 CFU in the public area. Meanwhile, the settled bacterial levels in both areas remained below 10 CFU (Figures 1B–1D). The buffer room met the Class 10000 cleanliness standard, with a particulate count (≥0.5 μm) < 350,000/m3 (Figure 1E). Overall, the significantly lower airborne/settled bacteria and particulate counts in the buffer room confirm its superior air cleanliness compared to the public area.Figure 1Status of airborne and settled bacteria, as well as dust particles, within the purification buffer room and the public area(A–E) Structural schematic of the PIVAS in our hospital: layout of the purification buffer room; (B, C, and D) Statistical analysis of airborne and settled bacteria, accompanied by their corresponding histograms; (E) Levels of particulate matter in the purified buffer room and public area subsequent to disinfection in our hospital, respectively. These outcomes are presented as the mean ± standard deviation (SD) (n = 3). ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001 when compared to the control group. Statistical significance was ascertained through the utilization of an unpaired Student’s t-test. Status of airborne and settled bacteria, as well as dust particles, within the purification buffer room and the public area (A–E) Structural schematic of the PIVAS in our hospital: layout of the purification buffer room; (B, C, and D) Statistical analysis of airborne and settled bacteria, accompanied by their corresponding histograms; (E) Levels of particulate matter in the purified buffer room and public area subsequent to disinfection in our hospital, respectively. These outcomes are presented as the mean ± standard deviation (SD) (n = 3). ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001 when compared to the control group. Statistical significance was ascertained through the utilization of an unpaired Student’s t-test. The optimal action concentration of ozone disinfection in PIVAS In the concentration-dependent single-factor experiment, the microbial loads of airborne, settled, and surface bacteria in the clean buffer room were quantified to evaluate the disinfection efficacy. As shown in Figure 2A, the inhibition of microbial growth exhibited a clear concentration-dependent pattern: Higher ozone concentrations led to stronger suppression of all three microbial indicators. The airborne bacterial count was reduced to <10 CFU at 7 ppm ozone and was nearly undetectable at 9 ppm. Concomitantly, the settled bacterial count decreased significantly at 7 ppm.Figure 2Status of airborne, settled, and surface bacteria under different concentration conditions in the purified buffer room exposed to varying ozone concentrations for 1 h(A–D) Statistical analysis of colony counts of airborne and settled bacteria, accompanied by their corresponding histograms; (D) Statistical analysis of colony counts of surface bacteria in the purification buffer chamber (a: rubber ground; b: glass; c: medical panel; d: door handle; e: pass-through window; I: PVC drug dispensing baskets; II: stainless-steel trolleys; III: Infusion bag; IV: ampoule). Status of airborne, settled, and surface bacteria under different concentration conditions in the purified buffer room exposed to varying ozone concentrations for 1 h (A–D) Statistical analysis of colony counts of airborne and settled bacteria, accompanied by their corresponding histograms; (D) Statistical analysis of colony counts of surface bacteria in the purification buffer chamber (a: rubber ground; b: glass; c: medical panel; d: door handle; e: pass-through window; I: PVC drug dispensing baskets; II: stainless-steel trolleys; III: Infusion bag; IV: ampoule). For the surface bacteria assessment, sampling was conducted at representative sites: rubber ground, glass surfaces, medical-grade panels, door handles, pass-through windows, infusion bags, stainless steel utility carts, polyvinyl chloride (PVC) drug dispensing baskets, and ampoules. Detectable bacterial colonies were primarily localized in rubber ground (a), PVC drug dispensing baskets (I), and stainless-steel trolleys (II). The colony counts decreased progressively with the increasing ozone concentration, reaching undetectable levels at 7 ppm (Figure 2D). Negligible bacterial growth was observed at all other sampling sites. Therefore, the ozone concentration was fixed at 7 ppm for the subsequent time-variable single-factor experiments. The optimal action time of ozone disinfection in PIVAS To determine the optimal ozone disinfection duration for PIVAS, 7 ppm ozone was applied across six exposure intervals (15–90 min, with 15-min increments). Microbiological efficacy was evaluated via quantitative analysis of airborne, settled, and surface bacteria in the purification buffer room. As shown in Figure 3A, airborne and settled bacteria exhibited significant time-dependent reductions, with logarithmic CFU decreases proportional to the exposure duration. Persistent microbial colonization was exclusively observed in airborne (3.7 ± 0.2 log CFU/cm2 at 15 min) and settled bacteria (0.4 ± 0.2 log CFU/cm2 at 15 min), where ozone induced progressive inactivation (80% reduction by 60 min, 96.2% by 90 min).Figure 3Status of airborne, settled, and surface bacteria in the purified buffer room under the influence of temporal variables subsequent to exposure to 7 ppm ozone(A–D) Statistical data on the colony counts of airborne and settled bacteria, accompanied by their corresponding histograms; (D) Statistical data on the colony counts of surface bacteria (a: rubber ground; b: glass; c: medical panel; d: door handle; e: pass-through window; I: PVC drug dispensing baskets; II: stainless-steel trolleys; III: Infusion bag; IV: ampoule). Status of airborne, settled, and surface bacteria in the purified buffer room under the influence of temporal variables subsequent to exposure to 7 ppm ozone (A–D) Statistical data on the colony counts of airborne and settled bacteria, accompanied by their corresponding histograms; (D) Statistical data on the colony counts of surface bacteria (a: rubber ground; b: glass; c: medical panel; d: door handle; e: pass-through window; I: PVC drug dispensing baskets; II: stainless-steel trolleys; III: Infusion bag; IV: ampoule). Surface bacteria (Figure 3D) at nine anatomical sites—including high-touch areas (rubber ground, PVC drug dispensing baskets, and stainless-steel trolleys) and low-contact surfaces (glass, medical panels, infusion bags)—revealed heterogeneous contamination. High-contact surfaces (≤1 log CFU/cm2 baseline) achieved <0.05 log CFU/cm2 after 60 min, demonstrating rapid microbial clearance. Response surface test design and verification test to determine the optimal parameters for ozone disinfection Based on previous single-factor experiments, a response surface test was designed (see Table S1). For ozone concentration (1–9 PPM) and exposure time (15–90 min), Design Expert 13.0 software with Central Composite design was employed. Log reduction (KL) values of natural bacteria in the airborne, settled, and surface bacteria on high-touch areas (rubber ground, PVC drug dispensing baskets, and stainless-steel trolleys) were calculated under different combinations to explore interaction effects and establish a disinfection model. To verify model prediction accuracy, four additional random points within the test range (not involved in model construction, Table S2) were selected for correlation testing between actual and predicted values (y = x). Quantitative modeling of dose-time dependency in ozone disinfection of airborne bacteria Multivariate analysis of variance (MANOVA) combined with quadratic response surface modeling was employed to quantify the ozone sterilization efficacy against airborne bacteria. The derived quadratic regression equation is: KL = −0.258297 + 0.104222X1 + 0.010535X2 + 0.001319X1X2 - 0.008198X12 - 0.000101X22, where (X1) denotes the ozone concentration (ppm) and (X2) represents the exposure time (min). Analysis of variance (Table 1) confirmed the exceptional significance of the model (F = 283.49, p < 0.001) with non-significant lack-of-fit (F = 1.31, p = 0.3872), indicating a robust fit without systematic deviation. The coefficient of determination (R2 = 0.9951 and adjusted R2 = 0.9916) demonstrated that 99.16% of the response variability could be explained, validating the predictive accuracy.Table 1Variance analysis results of the regression model for airborne bacteriasourcesum of squaresdfMean squaresF-valuep valueModel1.250.2403283.49<0.0001X1-Concentration0.63910.639753.82<0.0001X2-Time0.350910.3509413.98<0.0001X1X20.100210.1002118.18<0.0001X120.047510.047556.060.0001X220.022710.022726.790.0013Residual0.005970.0008––Lack of Fit0.006930.00231.310.3872Pure Error0.00140.0003––Cor Total1.2112–––R2 = 0.9951Radj2 = 0.9916predicted R2 = 0.9572C.V.% = 5.57Adeq precision = 57.4573 Table 1Variance analysis results of the regression model for airborne bacteriasourcesum of squaresdfMean squaresF-valuep valueModel1.250.2403283.49<0.0001X1-Concentration0.63910.639753.82<0.0001X2-Time0.350910.3509413.98<0.0001X1X20.100210.1002118.18<0.0001X120.047510.047556.060.0001X220.022710.022726.790.0013Residual0.005970.0008––Lack of Fit0.006930.00231.310.3872Pure Error0.00140.0003––Cor Total1.2112–––R2 = 0.9951Radj2 = 0.9916predicted R2 = 0.9572C.V.% = 5.57Adeq precision = 57.4573 Variance analysis results of the regression model for airborne bacteria Response surface analysis (Figure 4) revealed a significant dose-dependent efficacy for the ozone concentration (p < 0.0001), which was counterbalanced by concentration saturation (p < 0.05). The exposure time showed a strong time-dependent positive correlation (p < 0.0001) with diminishing marginal returns (p < 0.001). A significant interaction (X1X2, p < 0.0001) indicated a concentration-time synergy, with the sterilization efficacy (KL > 1.0) plateauing at ≥ 6 ppm and ≥60 min. Rigorous RSM uncovered the nonlinear interplay between the ozone concentration and the exposure time.Figure 4Response surface analysis of the inactivation efficiency of airborne bacteria with respect to ozone concentration and exposure time(A) Three-dimensional response surface plots of ozone concentration and exposure time.(B) Two-dimensional contours of ozone concentration and exposure time.(C) Design-expert plot of predicted versus actual values for the logarithm of bacterial inactivation.(D) Linear regression plot of four additional random points within the test range (not included in model construction). Response surface analysis of the inactivation efficiency of airborne bacteria with respect to ozone concentration and exposure time (A) Three-dimensional response surface plots of ozone concentration and exposure time. (B) Two-dimensional contours of ozone concentration and exposure time. (C) Design-expert plot of predicted versus actual values for the logarithm of bacterial inactivation. (D) Linear regression plot of four additional random points within the test range (not included in model construction). Dose-time kinetics of ozone sterilization against settled bacteria To characterize the ozone sterilization efficacy against passively sampled microbial colonies (settled bacteria), a quadratic regression model was developed through ANOVA and multiple regression analysis (Table 2). The finalized equation: KL = −0.314824 + 0.244533X1 + 0.004375X2 + 0.000259X1X2 - 0.015005X12 - 0.000019X22 (where X1 = ozone concentration [ppm], X2 = exposure time [min]) demonstrated robust statistical validity: F = 134.68 (p < 0.0001), non-significant lack-of-fit (F = 1.13, p = 0.436), and high explanatory power (R2 = 98.24%), indicating minimal experimental noise and strong congruence between the model and the data.Table 2Variance analysis results of the regression model for settled bacteriasourcesum of squaresdfMean squaresF-valuep valueModel1.4250.2832134.68<0.0001X1-Concentration1.0111.01480.11<0.0001X2-Time0.089310.089342.480.0003X1X20.004910.00492.330.0007X120.25510.255121.28<0.0001X220.000810.00080.39520.5495Residual0.014770.0021––Lack of Fit0.006830.00231.130.4356Pure Error0.00840.002––Cor Total1.4312–––R2 = 0.9897Radj2 = 0.9824predicted R2 = 0.9441C.V.% = 7.25Adeq precision = 34.1673 Table 2Variance analysis results of the regression model for settled bacteriasourcesum of squaresdfMean squaresF-valuep valueModel1.4250.2832134.68<0.0001X1-Concentration1.0111.01480.11<0.0001X2-Time0.089310.089342.480.0003X1X20.004910.00492.330.0007X120.25510.255121.28<0.0001X220.000810.00080.39520.5495Residual0.014770.0021––Lack of Fit0.006830.00231.130.4356Pure Error0.00840.002––Cor Total1.4312–––R2 = 0.9897Radj2 = 0.9824predicted R2 = 0.9441C.V.% = 7.25Adeq precision = 34.1673 Variance analysis results of the regression model for settled bacteria Response surface analysis (Figure 5) within the operational range revealed nuanced dose-time interactions. The sterilization efficacy showed a significant positive linear correlation with ozone concentration (p < 0.0001) but a negative quadratic dependence (p < 0.0001), reflecting saturation effects at higher doses. Time exerted extremely significant positive linear (p < 0.0001) and negative quadratic (p < 0.0001) influences, suggesting rapid initial inactivation followed by a kinetic deceleration. The concentration-time interaction term (p < 0.05) emphasized synergistic effects, with optimal sterilization achieved at 7–8 ppm for 60–75 min (the inflection point of the contour map).Figure 5Response surface analysis of the inactivation efficiency of settled bacteria by ozone concentration and exposure time(A) Three-dimensional response surface plots of ozone concentration and exposure time.(B) Two-dimensional contours of ozone concentration and exposure time.(C) Design-expert plot of predicted versus actual values for the logarithm of bacterial inactivation.(D) Linear regression plot of four additional random points within the test range (not included in model construction). Response surface analysis of the inactivation efficiency of settled bacteria by ozone concentration and exposure time (A) Three-dimensional response surface plots of ozone concentration and exposure time. (B) Two-dimensional contours of ozone concentration and exposure time. (C) Design-expert plot of predicted versus actual values for the logarithm of bacterial inactivation. (D) Linear regression plot of four additional random points within the test range (not included in model construction). Dose-time efficacy of ozone disinfection on surface bacteria Material-specific regression analysis via single-factor experiments characterized microbial inactivation kinetics on representative surfaces in a pharmaceutical buffer room. Microbiological loads were detected exclusively on rubber ground, PVC drug dispensing baskets, and stainless-steel trolleys, with viable counts exhibiting significant dose-dependent responses to ozone concentration (X1, ppm) and exposure duration (X2, min) as depicted in Figure 6.Figure 6Response surface analysis of the inactivation efficiency of surface bacteria with respect to ozone concentration and exposure time(A) Response surface plots for rubber ground.(B) Response surface plots for stainless steels.(C) Response surface plots for PVC drug dispensing baskets.(D) Linear regression plots of four additional random points within the test range (not included in model construction) for rubber ground, stainless steel, and PVC drug dispensing baskets, respectively. Response surface analysis of the inactivation efficiency of surface bacteria with respect to ozone concentration and exposure time (A) Response surface plots for rubber ground. (B) Response surface plots for stainless steels. (C) Response surface plots for PVC drug dispensing baskets. (D) Linear regression plots of four additional random points within the test range (not included in model construction) for rubber ground, stainless steel, and PVC drug dispensing baskets, respectively. For rubber ground, the validated regression model (R2 = 0.982, p < 0.001) revealed: KL = −0.2793 + 0.2028X1 + 0.00256X2 + 0.00199X1X2 - 0.0161X12 - 0.000023X22. Notably, the negative quadratic term for X1 (−0.0161X12) indicated saturation at ozone concentrations >6 ppm, consistent with rubber’s ozone-resistant polymer structure. Stainless-steel trolleys exhibited a distinct kinetic profile (R2 = 0.9945, p < 0.0001): KL = −0.2235 + 0.0887X1 + 0.00781X2 + 0.00168X1X2 - 0.00681X12 - 0.000073X22. The higher X2 coefficient (0.00781 vs. 0.00256) highlighted time-dependent inactivation dominance on non-porous surfaces. PVC plastic trays, sharing similar surface energy with stainless steel, yielded an analogous model (R2 = 0.9967, p < 0.0001): KL = −0.2317 + 0.0901X1 + 0.00799X2 + 0.00167X1X2 - 0.00656X12 - 0.000075X22. ANOVA results (Tables 3, 4, and 5) confirmed all models met statistical significance (p < 0.01 for all terms), with non-significant lack-of-fit tests (p > 0.05) validating predictive capabilities. Mechanistically, positive interaction terms [(0.00167, 0.00168, 0.00199)X1X2] indicated concentration-time synergy, aligning with ozone’s biphasic action: rapid oxidation of extracellular bio-molecules followed by slower cell wall penetration.Table 3Variance analysis results of the regression model for the rubber ground surfacesourcesum of squaresdfMean squaresF-valuep valueModel2.6850.5368215.33<0.0001X1-Concentration1.6511.65659.86<0.0001X2-Time0.582410.5824233.61<0.0001X1X20.22710.22791.03<0.0001X120.183710.183773.69<0.0001X220.001210.00120.4890.5069Residual0.017570.0025––Lack of Fit0.012530.00423.340.1372Pure Error0.00540.0012––Cor Total2.712–––R2 = 0.9935Radj2 = 0.9889Predicted R2 = 0.9503C.V.% = 6.96Adeq precision = 49.2425Table 4Variance analysis results of the regression model for the stainless steel surfacesourcesum of squaresdfMean squaresF-valuep valueModel1.6250.3241252.15<0.0001X1-Concentration0.885610.8856689<0.0001X2-Time0.503610.5036391.78<0.0001X1X20.161710.1617125.82<0.0001X120.032710.032725.480.0015X220.011810.01189.170.0191Residual0.00970.0013––Lack of Fit0.006930.00234.520.0894Pure Error0.00240.0005––Cor Total1.6312–––R2 = 0.9945Radj2 = 0.9905Predicted R2 = 0.9570C.V.% = 6.51Adeq precision = 55.3358Table 5Variance analysis results of the regression model for the PVC plastic surfacesourcesum of squaresdfMean squaresF-valuep valueModel1.6850.3366425.81<0.0001X1-Concentration0.956710.95671210.12<0.0001X2-Time0.497510.4975629.26<0.0001X1X20.161310.1613204<0.0001X120.030410.030438.430.0004X220.012610.012615.960.0052Residual0.005570.0008––Lack of Fit0.004430.00154.950.0781Pure Error0.001240.0003––Cor Total1.6912–––R2 = 0.9967Radj2 = 0.9944Predicted R2 = 0.9727C.V.% = 5.03Adeq precision = 71.9569 Table 3Variance analysis results of the regression model for the rubber ground surfacesourcesum of squaresdfMean squaresF-valuep valueModel2.6850.5368215.33<0.0001X1-Concentration1.6511.65659.86<0.0001X2-Time0.582410.5824233.61<0.0001X1X20.22710.22791.03<0.0001X120.183710.183773.69<0.0001X220.001210.00120.4890.5069Residual0.017570.0025––Lack of Fit0.012530.00423.340.1372Pure Error0.00540.0012––Cor Total2.712–––R2 = 0.9935Radj2 = 0.9889Predicted R2 = 0.9503C.V.% = 6.96Adeq precision = 49.2425 Variance analysis results of the regression model for the rubber ground surface Table 4Variance analysis results of the regression model for the stainless steel surfacesourcesum of squaresdfMean squaresF-valuep valueModel1.6250.3241252.15<0.0001X1-Concentration0.885610.8856689<0.0001X2-Time0.503610.5036391.78<0.0001X1X20.161710.1617125.82<0.0001X120.032710.032725.480.0015X220.011810.01189.170.0191Residual0.00970.0013––Lack of Fit0.006930.00234.520.0894Pure Error0.00240.0005––Cor Total1.6312–––R2 = 0.9945Radj2 = 0.9905Predicted R2 = 0.9570C.V.% = 6.51Adeq precision = 55.3358 Variance analysis results of the regression model for the stainless steel surface Table 5Variance analysis results of the regression model for the PVC plastic surfacesourcesum of squaresdfMean squaresF-valuep valueModel1.6850.3366425.81<0.0001X1-Concentration0.956710.95671210.12<0.0001X2-Time0.497510.4975629.26<0.0001X1X20.161310.1613204<0.0001X120.030410.030438.430.0004X220.012610.012615.960.0052Residual0.005570.0008––Lack of Fit0.004430.00154.950.0781Pure Error0.001240.0003––Cor Total1.6912–––R2 = 0.9967Radj2 = 0.9944Predicted R2 = 0.9727C.V.% = 5.03Adeq precision = 71.9569 Variance analysis results of the regression model for the PVC plastic surface Determination and verification of optimal ozone disinfection parameters for buffer room This study employed Design-Expert 13.0 to optimize ozone concentration (X1) and disinfection time (X2) within predefined ranges, aiming to maximize the sterilization response. Multiple near-optimal solutions (Table S8) resulted in non-integer parameters. Although these parameters are theoretically optimal, integer values are required for practical implementation. To address this constraint, a LINGO 18.0 integer programming model was developed. It restricted X1 (1–9 ppm) and X2 (15–90 min) to integers. The objective function was to minimize the operational input (X1 + X2) while maintaining at least 95% of the maximum response predicted by Design-Expert, thus aligning numerical optimization with real-world feasibility. LINGO optimization produced the optimal integer solution: X1 = 8 ppm and X2 = 67 min. This combination meets the operational requirements with minimal deviation from the theoretical optimum (≤1.2% response reduction, validated by simulation) and lies within the equipment tolerance limits (8 ppm: standard industrial concentration; 67 min: achievable via commercial timers). This workflow demonstrates a rigorous translation of computational optimization into practical application, resolving the engineering/biomedical challenge of balancing mathematical precision with field implementability. Integrating design-expert (RSM) and LINGO (integer programming) ensures scientific rigor and operational viability, providing a framework adaptable to similar parametric optimization problems. Evaluation of the bactericidal effect of ozone on common carrier materials in PIVAS and the sensitivity of bacterial strains Building on optimized ozone parameters (8 ppm, 67 min) obtained from experimental and simulation-based refinement, this study evaluated the antimicrobial efficacy of ozone against Staphylococcus aureus and Escherichia coli, which are prevalent PIVAS pathogens. It also assessed its compatibility with typical carrier materials (rubber, PVC, stainless steel). Controlled laboratory chamber testing under operational conditions demonstrated strong antimicrobial efficacy against both species, with no significant material-dependent differences. Notably, all surfaces (rubber, PVC, stainless steel) exhibited KL values >1.2, indicating consistent inactivation performance, as shown in Figure 7. These findings confirm the reliability of ozone as a cross-material disinfectant in PIVAS settings, achieving uniform effectiveness against target pathogens and common infrastructure materials.Figure 7Bactericidal efficacy of ozone-optimized parameters against common bacteria in PIVAS(A) Disinfection effect of ozone on E. coli between experimental and control groups, along with their histograms.(B) Disinfection effect of ozone on S. aureus between experimental and control groups, along with their histograms. (RG: rubber ground; PVC: polyvinyl chloride drug dispensing baskets; SS: stainless steel). Bactericidal efficacy of ozone-optimized parameters against common bacteria in PIVAS (A) Disinfection effect of ozone on E. coli between experimental and control groups, along with their histograms. (B) Disinfection effect of ozone on S. aureus between experimental and control groups, along with their histograms. (RG: rubber ground; PVC: polyvinyl chloride drug dispensing baskets; SS: stainless steel). Discussion This study developed an innovative ozone disinfection system integrated with the HVAC system for the buffer zone of PIVAS to address the inherent limitations of conventional disinfection protocols.15,16 Through a systematic approach that combines single-factor experiments, RSM modeling, and LINGO constraint optimization, the optimal disinfection parameters were identified as an ozone concentration of 8 ppm with a 67-minute exposure. Under these conditions, the KL values for S. aureus and E. coli exceeded 1.2. The airborne bacterial count was reduced to 6 CFU/m3, the surface bacterial count was ≤2 CFU/25 cm2, and the sedimentary bacterial count was ≤3 CFU/Plate, fully complying with China’s National Health Commission (NHC) PIVAS Construction Guidelines and ISO Class 7 environmental standards specified in USP <797>.6,17,18 Notably, ozone demonstrated excellent compatibility with common PIVAS materials (PVC, stainless steel, rubber), showing no observable corrosion or aging after long-term exposure, which further supports its clinical applicability and long-term operational feasibility.19 Additionally, post-disinfection ozone elimination via the HVAC system reduces the concentrations below the human exposure threshold (<0.15 ppm) within 5 min. The core innovation of this study lies in the integration of “technical design optimization” and “parameter precision calibration.” The integration of the ozone generator with the existing HVAC system resolved the uneven concentration distribution, which is a critical limitation of traditional ozone disinfection.19 This uniform distribution ensures consistent disinfection efficacy throughout the buffer room, eliminating the disinfection blind spots common in conventional systems. Furthermore, the application of RSM combined with LINGO constraint optimization achieves dual optimization: maximizing disinfection efficiency while minimizing operational costs and material damage.20 This dual-optimization approach outperforms the single-factor optimization methods commonly used in previous PIVAS disinfection studies, providing a more scientific and operable technical solution for sterile environment management in PIVAS. Mechanistically, RSM models (for airborne, settled, and surface bacteria) revealed a nonlinear interplay between ozone concentration and exposure time, featuring significant synergistic effects (X1X2 interaction term, p < 0.0001) and concentration saturation effects (negative quadratic terms for X1, p < 0.05).21 These findings deepen the understanding of ozone disinfection kinetics in PIVAS settings, advancing from empirical parameter selection to a data-driven, mechanistically informed optimization framework.22 To highlight the significance of this study and contextualize it within the existing literature, three pivotal works in ozone disinfection research were analyzed: Alimohammadi and Naderi demonstrated the efficacy of ozone in inactivating airborne viruses in enclosed spaces, providing direct evidence to support the extension of ozone disinfection to PIVAS environments.10 Ding et al. reported the high efficiency of ozone against chlorine-resistant bacteria in drinking water, confirming its broad-spectrum antimicrobial activity, which is a critical advantage for PIVAS applications.23 Cai et al. reviewed the oxidative inactivation mechanisms of ozone against waterborne pathogens, which are in line with our observations of bacterial inactivation through the oxidation of proteins, lipids, and nucleic acids.24 Building on these foundational studies, our work advances current knowledge by focusing on the integration of HVAC systems and the optimization of PIVAS-specific parameters. It delivers a practical and scalable disinfection solution for sterile compounding environments and addresses a critical gap in the field. Compared with conventional disinfection methods, ozone disinfection offers inherent mechanistic advantages. Chlorine-containing disinfectants denature proteins through hypochlorous acid, yet they induce the corrosion of PIVAS metal equipment and leave toxic residues. Conversely,25,26 UV irradiation only inactivates surface microorganisms on irradiated objects and fails to target pathogens in shadowed areas.27 In contrast, ozone decomposes into oxygen after disinfection without leaving harmful residues.28 This property aligns with the “green pharmacy” concept in global pharmaceutical practice and resolves the long-standing trade-off between efficacy and safety in PIVAS disinfection.29 Additionally, in the construction of PIVAS, ozone corrosion behavior exhibits significant material-specific characteristics between two critical materials: PVC and stainless steel. Experimental data confirm that these two primary construction materials have distinct corrosion mechanisms and tolerance thresholds.19 For PVC components, studies have reported that the material is “highly suitable” for ozone exposure, demonstrating excellent compatibility under gaseous ozone conditions, with no measurable degradation observed even after long-term exposure. In ozone aqueous environments, PVC maintains structural integrity at continuous concentrations of 0.2–0.5 ppm or short-term exposure to 1 ppm ozone. Tests involving immersion in high-purity deionized water containing 50 ppm dissolved ozone for 5 weeks showed only slight dehydrochlorination reactions without matrix damage.30 As the preferred metallic material in PIVAS systems, stainless steel forms a stable passive oxide film on its surface, exhibiting complete corrosion resistance in dry/pure gaseous ozone environments. However, in chlorinated ozone aqueous solutions typical of PIVAS disinfection processes, synergistic corrosion occurs with Cl− ions. In artificial seawater containing 0.2–0.4 mg/L dissolved ozone, the crevice corrosion penetration rate of 316L stainless steel is 5 mm/year, significantly lower than the 8 mm/year of 304 stainless steel.31 Notably, pitting or intergranular corrosion does not occur in chloride-free ozone aqueous solutions or pure gaseous environments. In summary, PVC demonstrates reliable corrosion resistance in gaseous ozone and temporary tolerance to low-concentration (≤0.5 ppm continuous) ozone water with minimal additive leaching. Whereas stainless steel ensures long-term stability in dry ozone systems but requires monitoring of chloride ion concentration (target <0.1 mg/L) in aqueous disinfection applications to prevent pitting corrosion.19,32 Therefore, both materials are suitable for PIVAS construction when operated within their specific ozone exposure parameter ranges. Although the present study only validated the ozone disinfection efficacy against E. coli and S. aureus, ozone, as a broad-spectrum microbicide, exhibits distinct inactivation efficiencies against microorganisms due to structural variations in their biological compositions. The concentration and exposure time required for ozone to inactivate bacteria vary significantly with the strain type and physiological state. Vegetative cells require lower concentrations and shorter treatment times, whereas spores necessitate higher concentrations or prolonged exposure. For instance, Chen et al. demonstrated that a 60-second exposure to 3.9 ppm ozone achieved a log reduction exceeding 5.9 for vegetative cells of Salmonella typhimurium and E. coli.33 Akbas et al. reported that applying 1.0 ppm gaseous ozone for the same duration reduced the Bacillus cereus vegetative cell counts by 3.5 log units, while the spores only showed a 2-log-unit reduction.34 Huang et al. established an ozone sensitivity hierarchy. Clostridium perfringens (with a 100–200 nm cell wall) exhibits partial resistance due to the cell wall thickness, and Penicillium citrinum spores with closed capsule structures demonstrate higher resistance than vegetative cells. Additionally, 8.64 ppm ozone for 15 min achieved 99.84% inactivation of E. coli, while 83 ppm ozone required only 5 min to achieve comparable efficacy.35 For fungal inactivation, ozone concentration and exposure time must be adjusted according to fungal species and morphological characteristics. Yeasts are more sensitive to ozone, while mold spores require higher concentrations or longer treatment durations. Notably, increasing the ozone concentration and prolonging the exposure time significantly enhances the inactivation efficacy. For example, Zhang et al. demonstrated that a 10 mg/L ozone aqueous suspension applied for 3 min achieved 100% inactivation of Aspergillus niger spores.36 Liu et al. reported that a 170 mg/m3 ozone treatment resulted in 58.5% and 73% inhibition rates against spores of Penicillium expansum and Botrytis cinerea (kiwifruit pathogenic fungi), respectively.37 Hu et al. showed that a 240 mg/m3 ozone exposure for 15 min achieved 28.25%–52% inactivation rates against four yeast species, including Candida albicans and Candida glabrata. Extending the treatment time to 45 min increased the inactivation rates to 66.57%–88.91%, and further increasing the ozone concentration to 2736 mg/m3 resulted in 100% inactivation of these four Candida species within 45 min.38 These data collectively indicate that ozone’s fungicidal efficacy improves with increasing concentration and treatment duration. Ozone exhibits higher overall efficiency in virus inactivation, requiring lower concentrations and shorter exposure times. Enveloped viruses are more sensitive than non-enveloped viruses, while single-stranded nucleic acid viruses show lower tolerance compared to their double-stranded counterparts. For example, Shin et al. demonstrated that a 0.37 mg/L ozone solution applied for 10 s achieved over 3 log units of norovirus inactivation.39 In another study, Brié et al. obtained similar results by exposing murine norovirus adsorbed on raspberries to 3 ppm gaseous ozone for 1 min, resulting in over 3.3 log units of virus inactivation.40 Huang et al. found that 5.3 ppm ozone for 45 min achieved 100% inactivation of hepatitis A virus, while 82.8 ppm ozone required only 1 min to achieve the same effect.35 These findings highlight the significant differences in ozone requirements for inactivating viruses compared to bacteria and fungi. Collectively, multiple studies indicate that microbial ozone sensitivity follows the descending order: enveloped viruses > non-enveloped viruses > Gram-positive bacteria > Gram-negative bacteria > fungi > spores > endospores. In conclusion, this study presents a scientifically validated ozone disinfection solution for the PIVAS buffer zone through innovative system design and precise parameter optimization. The findings not only expand the technical repertoire for PIVAS infection control but also align with global trends in green pharmacy and sterile compounding standardization. With further validation and refinement, this system exhibits substantial potential for widespread clinical application, thereby enhancing the safety and efficiency of IV medication preparation. Limitations of the study This study has several limitations that should be considered when interpreting the results and generalizing the findings. First, the research was conducted in a single hospital’s PIVAS center. This may limit the generalization of the optimized ozone disinfection parameters to other PIVAS facilities with different sizes, layouts, or HVAC system configurations. Second, the study focused on short-term disinfection efficacy. The long-term stability of the HVAC-integrated ozone disinfection system, including potential equipment wear and temporal changes in disinfection performance, was not evaluated. Third, microbial analysis mainly targeted common bacterial colonies. The inactivation efficacy of the ozone system against viruses, fungi, and drug-resistant pathogens remains inadequately investigated. Fourth, the potential impact of ozone residues on drug stability and long-term occupational health risks for PIVAS staff exposed to low-concentration ozone was not assessed. Finally, the optimization of disinfection parameters was based on laboratory-scale monitoring. The operational effectiveness of large-scale PIVAS buffer rooms requires further validation. Resource availability Lead contact Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Jing-Jing Ma (jingjingmajj@suda.edu.cn). Materials availability This study did not generate new unique reagents. Data and code availability •Data: All data are available in the article and in supplemental information and/or from the corresponding authors upon reasonable request.•Code: This article does not report original code.•Additional information: Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request. •Data: All data are available in the article and in supplemental information and/or from the corresponding authors upon reasonable request. Data: All data are available in the article and in supplemental information and/or from the corresponding authors upon reasonable request. •Code: This article does not report original code. Code: This article does not report original code. •Additional information: Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request. Additional information: Any additional information required to reanalyze the data reported in this article is available from the lead contact upon request. Acknowledgments This work was supported by the 10.13039/100014718National Natural Science Foundation of China (no. 82401353), the Suzhou Medical College-QiLu Medical Research Program of 10.13039/501100007824Soochow University (24QL200221), and the Research Projects of Suzhou Pharmaceutical Association (nos. Syhky202410 and SKYXD2022015). Author contributions Conceptualization, J.J.M. and J.A.B.; methodology, C.Y.Z., D.Y.S., and X.L.L.; investigation, C.Y.Z., X.L.L., and J.Q.Y.; writing – original draft, C.Y.Z.; writing – review and editing, C.Y.Z.; funding acquisition, C.Y.Z. and X.L.L.; resources, C.Y.Z.; supervision, C.Y.Z. and J.J.M. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERBacterial and virus strainsStaphylococcus aureusATCC 25923Wild typeEscherichia coliATCC 25922Wild typeSoftware and algorithmsDesign Expert 13.0Stat-Ease, Inc.Version 13.0SPSS 24.0IBM Corp.Version 24.0LINGO 18.0LINDO Systems, Inc.Version 18.0OtherOzone analyzerPulitong Ltd., P.R. ChinaPLT300-O3Portable ozone generatorYounousen Ltd., P.R. ChinaFL-Y-10Laser dust particle counterSanxing Clean Technology Co., Ltd., ChinaY09-310 REAGENT or RESOURCESOURCEIDENTIFIERBacterial and virus strainsStaphylococcus aureusATCC 25923Wild typeEscherichia coliATCC 25922Wild typeSoftware and algorithmsDesign Expert 13.0Stat-Ease, Inc.Version 13.0SPSS 24.0IBM Corp.Version 24.0LINGO 18.0LINDO Systems, Inc.Version 18.0OtherOzone analyzerPulitong Ltd., P.R. ChinaPLT300-O3Portable ozone generatorYounousen Ltd., P.R. ChinaFL-Y-10Laser dust particle counterSanxing Clean Technology Co., Ltd., ChinaY09-310 Experimental model and study participant details This study did not involve human participants, animals, plants, or genetically modified organisms. The bacterial strains used were standard reference strains. S. aureus ATCC 25923 (wild type) and E. coli ATCC 25922 (wild type) are wild - type strains and were maintained on nutrient agar at 37°C. Strain identities were verified through standard microbiological procedures, including morphological examination and biochemical testing. No cell lines were used; authentication and mycoplasma contamination testing are not applicable. Method details Assessment of ozone concentration in the buffer room Upon the activation of the ozone disinfection system, the ozone concentrations within the chamber exhibit temporal variability and spatial heterogeneity, which makes it necessary to conduct continuous concentration monitoring. During the ozone disinfection process in the buffer room, the real - time concentration dynamics were monitored using an ozone analyzer (Model PLT300 - 03, Pulitong Ltd., PR China), and the temporal concentration profiles were constructed based on continuous measurements. To determine the optimal ozone disinfection parameters for the PIVAS environment, a portable ozone generator (Model FL - Y - 10, Younousen Ltd., PR China) with adjustable output intensity and operation time was employed in the buffer room. After the completion of operational procedures and the evacuation of personnel, the generator was initiated. Once the ozone level reached and stabilized at the pre - defined target concentration, the disinfection process was sustained for the specified period. The disinfection effectiveness was evaluated by quantitatively assessing three crucial microbial indicators: airborne bacteria, settled bacteria, and surface microbial contamination. Response surface methodology (RSM) combined with validation experiments was utilized to ascertain the optimal ozone concentration and exposure duration for PIVAS disinfection. As illustrated in Figure S1, the equilibration durations necessary for the ozone generator to reach target concentrations of 1, 3, 5, 7, and 9 ppm were 15, 30, 45, 60, 75, and 90 minutes, respectively. Significantly, the equilibration period should be excluded from the total operational time when validating the effective disinfection durations in subsequent experimental processes. Monitoring of airborne particles Sampling and detection were performed subsequent to the completion of all operational procedures and evacuation of personnel from the site. Sampling points were uniformly distributed on a horizontal plane at a height of 0.8 meters above the floor. A laser dust particle counter (Y09-310, Sanxing Clean Technology Co., Ltd., China) was used to quantify the count of airborne particles ≥ a specified size per unit volume in the clean zone, thereby assessing the cleanliness level of the controlled environment. Each sampling point was measured continuously for 5 replicate readings. Detection of airborne bacteria Prior to sample collection, aseptic disinfection of the airborne bacteria sampler was carried out. The external surfaces and the sampling port were disinfected with 75% ethanol. Subsequently, the outer surface of 90 - mm - diameter Petri dishes was sterilized before they were fixed into the sampling port. The sampler was placed on the operating table at a distance of 0.8 meters from the air supply surface, and the sampling duration was programmed; the instrument would automatically terminate operation upon completion. After sample collection, the Petri dish was immediately removed and inverted in an incubator within 4 hours, followed by incubation at 37°C for 48 hours. Colony formation was observed, and blank control procedures were simultaneously implemented.(Equation 1)F=N1vt1×1000 ∗F is the total number of airborne planktonic colonies (CFU/m3), N1 is the average value of colonies on the Petri dish (CFU), ν is the sampling speed (L/min), and t1 is the sampling time (min). Detection of settled bacteria Following the completion of all operational procedures and the evacuation of personnel from the site, detection was carried out. Five sampling points were set up within the room, encompassing the four corners and the central area. Each corner sampling point was located at a distance of 1 meter from the wall. A standard nutrient agar plate with a diameter of 90 mm was placed at each sampling location. At the start of sampling, the plate lid was removed and placed upside - down beside the plate. After an exposure duration of 30 minutes, the lid was replaced, and the plate was promptly transferred to a 37°C incubator for a 48 - hour period to enumerate bacterial colonies.(Equation 2)C=N2S1t2 ∗C is the total number of airborne settled colonies (CFU/m2.min), N2 is the average value of colonies on the Petri dish (CFU), S1 is the collection area of the Petri dish (the sampling area is usually the bottom area of the Petri dish, that is, π×(0.09/2)2 ≈ 0.00636 m2), and t2 is the exposure time (min). Detection of bacterial colonies on the object surface Surface sampling was carried out using sterile cotton swabs at pre - determined detection points. An area of 10×10 cm2 was sampled through horizontal and vertical smearing five times, accompanied by the concurrent rotation of the swab. The hand - contacted part of the swab was aseptically excised prior to immersion into a 10 mL sterile elution tube containing the appropriate neutralizing agent. Samples were transported to the laboratory within 30 minutes after collection. After thorough vortex - mixing of the swab eluate, 200 μL of the suspension was inoculated onto nutrient agar plates. The inoculated plates were incubated at 37°C for 48 hours, after which the bacterial colonies were enumerated.(Equation 3)P=N3c10abS2 ∗P is the surface bacterial concentration (CFU/cm2), N3 is the average value of colonies on the Petri dish (CFU), c is the volume of the sample stock solution (mL), a is the sample conversion coefficient, b is the volume of the liquid coated on the plate (mL), and S2 is the sampling area (cm2). Carrier quantitative bactericidal experiment Bacterial suspensions and contaminated carriers of S. aureus (ATCC 25923) and E. coli (ATCC 25922) were prepared according to the protocols outlined in the “Disinfection Technical Specification” (2002 edition).1 (a) Vegetative bacterial cell suspension preparation: Cultures were harvested by rinsing the slant surface with fresh medium, followed by serial dilution to achieve the required bacterial concentration. (b) Bacterial spore suspension preparation: The bacterial suspension was inoculated onto nutrient agar in Roux flasks and incubated for 7 days. Spore formation was confirmed via staining, and qualified spores were collected to prepare the suspension for subsequent use. (c) Preparation of PIVAS carriers: Common PIVAS materials (rubber, stainless steel, and plastic) were cut into 12 mm diameter circular carriers. Three replicates of each material were surface-cleaned and sterilized. Bacterial vegetative cells or spore suspensions were diluted in tryptic soy broth (TSB), and 10.0 μL aliquots were evenly spread onto each carrier. After air-drying, carriers were stored at 4°C until use. The positive control group showed bacterial recovery of 1×106–5×106CFU per carrier. (d) Disinfection efficacy testing: Contaminated carriers were placed in sterile Petri dishes and arranged in the test chamber. The chamber door was sealed, and the ozone disinfection system was activated with continuous ozone concentration monitoring. Upon reaching the pre-specified concentration and exposure time, bacterial carriers were collected alongside untreated positive controls, and immediately transferred to sterile test tubes containing neutralizing agent for 10 minutes. After thorough vortex mixing for elution, 200 μL of the eluate was inoculated onto nutrient agar plates. Plates were incubated at 37°C for 24 hours, followed by colony enumeration and log reduction calculation. All experiments were performed in triplicate.(Equation 4)KL=log10(EbeforeEafter) ∗KL is the logarithm of bacterial killing (log10 CFU/cm2), Ebefore is the bacterial concentration before disinfection (CFU/cm2), and Eafter is the bacterial concentration after disinfection (CFU/cm2). Response surface experimental design and verification test Initially, the optimal level ranges of the two independent variables were ascertained through single - factor experiments. Subsequently, a Central - Composite Design (CCD) within the Response Surface Methodology (RSM) framework was utilized. The Design Expert 13.0 software (Stat - Ease, Inc., USA) was employed to design experiments encompassing two factors: ozone concentration (1 – 9 ppm) and exposure time (15 – 90 min). The logarithmic reduction of natural bacteria in the PIVAS buffer room was computed under different disinfection combinations to investigate the interactions between these variables and establish a disinfection model. To verify the predictive accuracy of the model, four additional points within the experimental interval were randomly chosen, and a correlation analysis was carried out between the observed and predicted values (y = x). After validating the model's validity, software - based model analysis was conducted to obtain optimal disinfection parameters, ultimately determining the ideal ozone concentration and exposure time for PIVAS disinfection. All experiments were conducted in triplicate, and the mean values were used for subsequent analyses. Quantification and statistical analysis All experiments were conducted with a minimum of three independent biological replicates. Data were presented as the mean ± standard error of the mean (SEM) and analyzed using SPSS version 24.0 (IBM Corp., Armonk, NY, USA). The Student's t - test was utilized for comparisons between two groups, whereas one - way analysis of variance (ANOVA) with Tukey's post - hoc test was applied for multiple group comparisons. Statistical significance was defined as ∗P < 0.05, ∗∗P < 0.01, and ∗∗∗P < 0.001 when compared to the Sham or Blank group. Sample sizes (n = 3) were determined based on preliminary experiments to guarantee sufficient statistical power. No data were excluded from the analysis.