Organic Process Research & Development Concept Article pubs.acs.org/OPRD Quality by Design in Action 1: Controlling Critical Quality Attributes of an Active Pharmaceutical Ingredient Abdul Qayum Mohammed/'* Phani Kiran Sunkari,1 P. Srinivas,*'5 and Amrendra Kumar Roy*' ^CTOTII, Dr. Reddy's Laboratories Ltd, Plot 116, 126C and Survey number 157, S.V. Co-operative Industrial Estate, IDA Bollaram, Jinnaram Mandal, Medak District, Telangana 502325, India ^Department of Chemistry, Osmania University, Hyderabad, Telangana 500007, India f let ABSTRACT: The importance of Quality by Design (QbD) is being realized gradually, as it is gaining popularity among the generic companies. However, the major hurdle faced by these industries is the lack of common guidelines or format for performing a risk-based assessment of the manufacturing process. This article tries to highlight a possible sequential pathway for performing QbD with the help of a case study. The main focus of this article is on the usage of failure mode and effect analysis (FMEA) as a tool for risk assessment, which helps in the identification of critical process parameters (CPPs) and critical material attributes (CMAs) and later on becomes the unbiased input for the design of experiments (DoE). In this case study, the DoE was helpful in establishing a risk-based relationship between critical quality attributes (CQAs) and CMAs/CPPs. Finally, a control strategy was established for all of the CPPs and CMAs, which in turn gave rise to a robust process during commercialization. It is noteworthy that FMEA was used twice during the QbD: initially to identify the CPPs and CMAs and subsequently after DoE completion to ascertain whether the risk due to CPPs and CMAs had decreased. ■ INTRODUCTION Nowadays, Quality by Design (QbD) has become an essential part of any process development pertaining to drug substances,1'2 drug products, and analytical method development.3 QbD is based on Juran's concept of "planning quality into the product"4 at the design stage itself, rather than "complying product to the quality or Quality by QC". Designing quality into the product can be achieved only by having a proper understanding of the relationship between the critical quality attributes (CQAs) and the critical process parameters (CPPs) and critical material attributes (CMAs), as shown in Figures 1 and 2. It is based on the Two ways of optimization Traditional way_j_QbD/6o way yiQTPP, CQÄ) = f{x), where x = CPP/CMA Target is to meetthe quality Variation in product quality Reprocessing Rework Cost of poor quality (COPQ) Quality by QC Figure 1. Two approaches for optimization. Controlling the process parameter affecting a quality parameter Consistent quality 4- Less Reprocessing Less Rework Reduction of wastages Quality by Design concept of quality risk management, where one needs to assess the risk of each of the process parameters (PPs) and material attributes (MAs) on the CQAs. Various QbD guidelines have been published by different regulatory agencies6 to ensure risk mitigation for patients and also to fulfill their own key Dependentvariable Response Effect Out-put characteristics Response Independent variable Factor/variable Cause Input Process variables Factor/variable ACS Publications e 2015 American Chemical Society If you know and control the process variables (X} then you can control your product quality (Y]. Thisis QbD. Figure 2. Relationship between CQAs and CPPs/CMAs. responsibility areas (KRAs) of acceptability, affordability, availability, and accessibility (known as the 4As).7 On the other hand, manufacturers are realizing that in addition to taking care of patient safety they also need to ensure that they adhere to the 4A's to retain their market share and to make some profits. One of the major hurdles to any robust process is inadequate understanding of the process, which results in inconsistent quality of the active pharmaceutical ingredient (API). Process robustness is achievable only through the implementation of proven risk-based statistical tools such as Six Sigma and QbD during the developmental and manufacturing phases of an API. Regulators have been flexible in accepting any quality risk tool Special Issue: Application of Design of Experiments to Process Development Received: September 15, 2014 Published: January 21, 2015 1634 DOI:10.102Vop500295a Org. Process Res. Dev. 2015,19,1634-1644 Organic Process Research & Development Concept Article Figure 3. Various sources of variation in a process. QbD API "API2 \ Traditional approach [am] 2+Oa2+Op2+0«2 Where a2 = variance Risk m Property 1 CtTPPs Assessment-l CQAS ■ Risk Assessments Input variables ID0E W CMAS& Design '-" Control [ Risk . forDoE 1 ■ CPPs Space Strategy Assessment-3 1 Control Strategy for all CPPs ^ J \ Feedback mechanism for continuous improvement Update FMEA-2 Document Safety * CQAs from single stage Figure 4. Sequence of steps for conducting QbD. that can mitigate the risk to a CQA caused by various process variables (CPPs/CMAs).8 There are many sources of variation in a process. The total variance in any CQA of an API is the sum of the individual variances contributed by all of the sources, as shown in Figure 3 and by eq 1: °API — "iCSM + "reagent + °CPP + "analysis + G' external factors (1) where denotes the variance of x. Additionally, the situation becomes much more complex if the process is a multistep synthesis, as represented by eq 2: j2 "step 1 72 "step 2 step n (2) There are certain sources of variation that can be controlled during process optimization, such as variations of the key starting material (KSM) and reagents (collectively called the CMAs) and the CPPs. These controllable variables are also known as assignable causes of variation. However, there are certain sources of variation that cannot be controlled (variations due to external factors such as room temperature, production shift, age of the reactors, operators, etc.), and these are collectively called common causes of variation.9 QbD helps in eliminating the assignable causes of variation by defining the range within which the CPPs/ CMAs can be varied. Nevertheless, the common causes of variation cannot be eliminated, and we have to live with some degree of inherent variability. However, QbD helps in minimizing the effect of these common causes of variation by randomization and blocking10 of experiments during the DoE study. Thus, QbD takes care of both types of variations and minimizes their effects on the CQAs of an API. Hence, QbD is a risk mitigation tool that ensures that the quality of an API produced in each batch remains the same, which in turn ensures that patient safety and the 4A's requirement of the regulators will be met. ■ PROCESS DEVELOPMENT OF AN API USING QBD The basic outline of the application of QbD in the process development of a drug substance is shown in Figure 4, and a detailed process is represented by the flow diagram in Figure 5. The Regulators allow the manufacturers to use any established risk analysis tools or any risk assessment tool developed in-house, as long as it serves the purpose. A possible sequence of events for QbD is described as follows (Figure 4): (l) categorization of drug properties; (2) identification of CQAs from quality target product profiles (QTPPs) (risk assessment l); (3) identification of CPPs and CMAs (risk assessment 2); (4) optimization of the effect of input variables; (5) control strategy; (6) re-evaluation of the risk to the CQAa (risk assessment 3); and (7) continuous improvement. These steps are discussed in more detail below. Step 1. Categorization of Drug Properties. All of the properties of the drug substance are placed into physicochemical, analytical, and safety categories, as all of these are treated separately. Step 2. Risk Assessment 1: Identification of CQAs from QTPPs. The risk assessment tool known as failure mode and 1635 DOI: 10.1021/op500295a Org. Process Res. Dev. 2015,19,1634-1644 Organic Process Research & Development Concept Article Figure 5. Process flow for QbD. Feedback for Continuous Improvement Process | ^> § Monitoring -USL LS L Improve Detecting assignable causes LSL Mean USL -USL UCL LCL LSL UCL UCL LCL UCL Improved Process Process not capable of meeting CQA specifications Process just capable of meeting CQA specifications Process is more than -capable of meeting CQA specifications Expected Process Improvement by QbD Figure 6. Ultimate goal of QbD. The CQA or the customer's requirement is represented by specification limits, whereas the process capability is represented by control limits. effect analysis (FMEA) is applied to screen the QTPPs in each of the above categories. This stage is important as it shortlists the QTPPs that are critical for the patients and must be in the API. These shortlisted QTPPs become the CQAs. This process is denoted as FMEA-1 in Figures 4 and 5. This stage requires due diligence for a new chemical entity (NCE), but for a generic molecule, it is the same as the specifications set by the customer or as given in the pharmacopeia. Step 3. Risk Assessment 2: Identification of CPPs and CMAs. Once the CQAs are identified, it becomes imperative to identify the PPs and MAs that can affect those CQAs. This is done by listing all of the PPs and MAs (or input variables for DoE, as shown in Figure 4) without any bias. Working backwards, i.e., listing the CQAs of the final API first, is recommended. This helps in identifying the origin of a particular CQA (e.g., an impurity), which in turn helps in controlling it at its point of origin (earlier stages). Once all of the PPs and MAs are listed, a second risk assessment (denoted as FMEA-2 in Figures 4 and 5) is performed to identify the important input variables from the PPs and MAs, either by brainstorming with the subject experts or by coming to a decision based on past experimental data. If this process is performed without bias, it reveals the important input variables that will become the input variables for the DoE. A tool such as FMEA, quality function deployment (QFD), or any tool developed in-house can be used for risk analysis, as the regulators are flexible about the choice of risk assessment tool. FMEA is the simplest of all the risk assessment tools. The input variables can be identified by using the following three criteria from FMEA analysis: a. unit operation with highest risk priority number (RPN) b. if there is no control strategy for any unit operation c. if the effect of any unit operation on the CQA is not known The FMEA template is shown in Table 2 with an example. 1636 DOI: 10.1021/op500295a Org. Process Res. Dev. 2015,19,1634-1644 Organic Process Research & Development Concept Article Scheme 1. Synthetic route to API hydrochloride 5" (CH2)n Ri ,0 N-R2 O O ^(CH2)n HN^< VnH ' O Hydrolyzed impurity 6 R!-N R, O (H2C)n^ O O r(CH2)n \_J HN-' Dimer impurity 7 Lactam impurity 8 "Reagents: (a) SOCI2, toluene; (b) potassium phthalimide, DMF/H20; (c) 40% aqueous methylamine solution; (d) EtOAc/HCl gas. Step 4. Optimization of the Effects of the Input Variables on the CQAs. On the basis of the screened input variables, a DoE is planned to gain an understanding of the relationship between the input variables and CQAs. The output of the DoE is the set of input variables that affect the CQA significantly and are termed as CPPs and CMAs. Additionally, the DoE also provides a design space within which CPPs and CMAs can be varied. It is possible that only a few of the input variables that were initially selected might actually affect the CQAs. Many times it happens that the same CPPs affect two or more CQAs, in which case all of the CQAs should be studied together by DoE or by using multivariate analysis (MVA) tools. The design space obtained from DoE provides an amicable region within which any CPP or CMA can be varied without affecting the CQAs. This becomes the basis for the control strategy. Step 5. Control Strategy.11 Once the operating ranges of the CPPs are identified (from the design space), it is important to ensure that all of the CPPs remain within their ranges by providing proper control during the manufacturing phase (e.g., by the use of process analytical tools (PATs) such as ReactIR, pH control, etc.). It is crucial that the manufacturers include their suppliers during the QbD phase to ensure that the manufacturers have control of the CMAs for all of the KSMs, as shown in eq 1. Another important substep of the control strategy is to monitor the CPPs along with the CQAs using individual-moving range (i-MR) control charts9 during commercialization in order to capture any deviations in the process. All of the deviations need to be investigated, and the reasons for positive deviations must be incorporated in the process, whereas the reasons for negative deviations must be eliminated. This enables the manufacturer to plan for continuous improvement during the entire life cycle of the product (Figure 6). As a general practice, only the CQAs are monitored by the control charts. However, one needs to understand that as the CQAs are the outcome of the CPPs (Figure 2), it is imperative to monitor both. Step 6. Risk Assessment 3: Risk Re-evaluation. After identification of the CPPs/CMAs and the design space and development of the control strategy, it is time for the third risk assessment (denoted as FMEA-3 in Figures 4 and 5), in which the risk to the CQAs is re-evaluated to determine whether it has been reduced after optimization with respect to the risk that existed during FMEA-2. There are many tools for risk assessment, but the most widely applied tool is PMEA. IfPMEA is applied in risk assessments 2 and 3, it becomes important to see whether the RPN of each CQA has decreased after risk assessment 3. If it has not, then one is working with the wrong CPPs/CMAs. Step 7. Continuous Improvement: Monitoring and Improving the Process. Even after all of the above steps have been performed, it is seldom observed that commercialization happens without any hiccups, as the process takes its own time to mature. The main reason that any CQA goes beyond the specification limits is the narrow gap between the customer's expectations and the capability of the process, as shown by Figure 6. For a given CQA, its specification is recommended on the basis of patient safety data, whereas the final CQA of any product is the outcome of the process, or in other words, the final specification of any CQA is determined by its process capability9 (Figure 6). If the process capability of any process is not under control, it would lead to out-of-specification (OOS) or out-of-trend (OOT) batches, which would trigger investigation. Some of the above deviations are good for the process (e.g., a yield increase) and some are bad (e.g., an increase in the impurity Scheme 2. Synthetic scheme for the conversion of 4 to 5 °Y^(CH2)n R^N^R2 ^NH2 EtOAc/lPA EtOAc.HCI (CH2)n N L + R{ "R2 NH3CI 5 API Hydrochloride 1637 DOI: 10.1021/op500295a Org. Process Res. Dev. 2015,19,1634-1644 Organic Process Research & Development Concept Article Specification of compound 3 Specification of Methylamine solution = CMA4for stage 4 CPP4 For Stage 4 This part of the work reported herein Specification of compound 4 Specification of EtOAc/HCI Use of QbD in optimizing the effect of CPP„ & CMA4 on CMA5 Subject of next publication = CMA5for stage 5 CPP5 For Stage 5 7 Specification of Final API (compound 5) CQA of API Figure 7. Various terms used in the current work and the companion article. CQAs of API (5) Acid alcohol impurity Lactone impurity Amide alcohol inipiirily £_ E <•> *o u "3 c5 a ~~ P. — — p C Lactam impurity 8 Specifications <0.15% <0.15% <0.15% .,Q 3X 5) is analyzed in this article to enable the readers to grasp the concept easily. The synthesis of the API hydrochloride (5) starts with the chlorination of amide alcohol 1 to give chloro compound 2, which upon substitution with potassium phthalimide results in protected amine 3. Deprotection of the phthalimide group with methylamine aqueous solution provides the API as a free base (4), which on treatment with EtOAc/HCI gives the final API as a hydrochloride salt (5). There were four major impurities (unreacted 3, hydrolyzed impurity 6, dimer impurity 7, and lactam impurity 8) that formed during the conversion of free base 4 to its hydrochloride salt 5, as shown in Scheme 1. The steps involved in the QbD as described above were applied to the salt formation stage of the API (5), as elaborated in the next section. Before the QbD is discussed, however, it is important to understand the terminologies used. The whole work was divided into two portions that are described in two Table 1. Identifying important MA5 for the manufacture of API 5 specifications maximum tolerable process control are these MA5 input for stage 5 limits limits'* important?^ remarks EtOAc/HCI NLT 896/8-12% 8-12% yes HC1 concentration to be in the range 8—12% compound 4 2.1. assay as per analysis as per analysis no taken to the next stage on the basis of the assay of 4 2.2. residual toluene as per analysis as per analysis no 2.3. unreacted 3 NMT 1% 0.5% yes even though 3, 6, and 7 would not participate in the next stage, it was 2.4. hydrolyzed impurity 6 NMT 3% 1.5% yes desired to keep them at minimum level 2.5. dimer impurity 7 NMT 3% 0.5% yes "These control limits were proposed after optimization of compound 4 using DoE. bThese important MA5 would be taken as input variables for the DoE studies. 1638 DOI: 10.1021/op500295a Org. Process Res. Dev. 2015,19,1634-1644 Organic Process Research & Development Concept Article Yield % Design-Expert® Software Factor Coding: Actual Yield % • Design Points ■ 87.4113 178.3717 X1 = A: IPA Volume X2 = C: HCl Mole Ratio. Eq 1.20 1.15 Actual Factor B: Cool Temp. ■■ 20.00 CT tu CC CC o o Ü x Ü 1.05 1.00 0.80 1.10 1.40 A: IPA Volume 1.70 Z.Ol Figure 9. Contour plot of the percent yield variation with respect to the volume of IPA and amount of HC1 at 20 °C. Design-Expert® Software Factor Coding: Actual Overlay Plot Yield % X1 = A: IPA Volume X2 = C: HCl Mole Ratio. Eq Actual Factor B: Cool Temp. = 18.78 Overlay Plot UJ o ra 0Ĺ 1.10 o O Ü Design Space for Yield > 80% Yield ±- so ono I 1.61 A: IPA Volume Figure 10. Design space. separate articles, as summarized in Figure 7. Herein the term CQA is only applied to the final API hydrochloride 5, or stage 5. These CQAs are affected by the critical material attributes and critical process parameters related to Stage 5, which are denoted as CMAS and CPPS, respectively. The CMAS concluded from the current work serve as the input for the work to be described in the companion article (DOI: 10.1021/ op500297g), wh ere the effects of CMA4 and CPP4 on CMAS are studied. Application of QbD to the Manufacture of API Hydrochloride 5. Step 1: Listing of All of the Quality Attributes Associated with the API. Since it is a generic molecule, all of the quality attributes are the same as the specification set by the customer with respect to the impurities, as shown in Figure 8. These quality attributes become the QTPPs. Step 2: Risk Assessment: Identification of the CQAs from the QTPPs. As stated earlier, since the API was a generic molecule, all of the QTPPs eventually became CQAs because all of them were 1639 DOI: 10.1021/op500295a Org. Process Res. Dev. 2015,19,1634-1644 Organic Process Research & Development Concept Article Table 2. FMEA analysis of the process used to manufacture final API hydrochloride 5 CQAs API (5) s. No activity potential failure mode Ol & hydrolyzed impurity 6 unreacted 4 impurity HCl content dimer impurity 7 lactam impurity 8 failure mode present control status of present control occurrence severity o s a o tj "5 ■o RPN 1 charge 8.5-10 volumes of EtOAc into the reactor more quantity ofEtOAc $ $ t $ t $ t measurement error calibrated day tank working 2 5 2 20 low quantity of EtOAc 4- $ t t $ $ t measurement error calibrated day tank working 2 9 2 3fi 2 charge 1-1.5 Volumes of 1PA into the reactor more quantity of IPA 4- $ t t $ t I measurement error calibrated day tank working but flow meter to 5 9 7 315 low quantity of IPA t 4- t t $ t I measurement error calibrated day tank be installed 5 9 7 315 3 charge crude stage 4 into the reactor more quantity of4 4- $ $ $ t analytical error toluene correction to the assay working 2 9 2 36 less quantity of 4 4- $ t $ I t I analytical error toluene correction to the assay working 2 9 2 36 4 stir the reaction mass till clear dissolution is observed what if clear dissolution not observed 4, $ t t t t t manual error not critical continue stirring till clear solution is obtained working 2 9 2 36 5 add EtOAc.HCl into the reaction mass at 25±10°C charging at high temperature 4, 4, t t I $ T failure of chilled water line valve of RT water not closed ensuring that RT water line is closed working 5 9 5 225 charging at low temperature 4* t t t $ $ $ RT water temperature fluctuation during winter no effect No control 5 7 7 245 charging of more eq. of IIC1 4- 4- t $ $ 0 t manual error releasing material 5 7 9 315 charging of less eq. of HCl 4- $ t t $ $ t on vendor's COA working 5 5 9 225 fast addition of EtOAc.HCl" 4- 4- t t $ $ t addition at constant 5 5 5 125 slow addtion of EtOAc.HCl I $ t t $ $ $ manual error rate with flow meters in 45-60 minutes working 5 5 5 125 more maintenance time t $ t $ $ $ $ no effect at 25+10°C not required not required 5 2 3 30 6 maintain the reaction mass at 25±10 °C for 2 hour less maintenance time 4- 4- t $ $ $ manual error IPC for the absence of 4 working 2 7 3 54 maintenance at high temperature 4- 4- t t t $ t failure of chilled water plant standby brine supply working 2 7 3 54 maintenance at low temperature t t t $ $ No effect Not required Not required 2 5 3 30 7 filter the compound under nitrogen atmosphere. filter under atmospheric conditions 4, t t t $ T Failure of nitrogen plant Standby nitrogen cylinders working 3 9 3 XI T increase in desired CQA T increase in undesired CQA $ no effect of CPPs on CQA •I decrease in undesired CQA I decrease in desired CQA critical to patient safety. Hence, in case of generic molecules, risk assessment to identify CQAs from QTPPs (FMEA-1 in Figures 4 and 5) is generally not required. The important point to note here is that the stage considered for QbD involved only salt formation (Scheme 2) whereas all of the impurities listed as CQAs (Figure 8) came from the previous 1640 DOI: 10.1021/op500295a Org. Process Res. Dev. 2015,19,1634-1644 Organic Process Research & Development Concept Article Table 3. Three important input PP5 considered for DoE studies on the basis of FMEA" variable [RPNb] levels used for DoE symbol A B C input variable for DoE volume of IP A addition temperature amount of HC1 low (-1) 0.8 volumes [315] 5 °C [245] 1 equiv [225] high (+i) 2 volumes [315] 35 °C [225] 1.2 equiv [315] "A 23 full factorial design with four center points was planned to study the effect of the three important input PP5 on the CQAs with all of the other PPs kept at predefined levels. bRisk priority numbers (RPNs) were taken from Table 2. s. no. PP limit justification 1 EtOAc volume 8—12 volumes no effect on the responses at high volume 2 time the reaction mass is stirred until clear dissolution not defined not critical if stirred for a longer time, as it is just for dissolution 3 addition time of EtOAc/HCl 45-60 min to control the exothermicity of reaction 4 cooling temperature before addition 10-15 °C a low temperature was chosen to control the exothermicity 5 maintenance temperature 10-15 °C same as the cooling temperature 6 maintenance time 2-2.5 h no effect on the responses after 1 h 7 filtration under nitrogen dry atmosphere required stages. On the basis of the laboratory data, all of those impurities at the final stage were washed away in mother liquors (MLs), as they could not form the HC1 salt. Figure 8 provides a summary of all of the CQAs, their points of origin, and their limits in the final API. Step 3: Identification of the important input Variables for the DoE Studies from All of the Material Attributes (MA^ and Process Parameters (PPs) Pertaining to Stage 5. Step 3.1: identification of important MAS. As compound 4 was the starting material for the final API 5, all of the attributes of compound 4 that could affect the CQAs were considered as MAS for the final API. This information was then used to identify important MAS along with their tolerable limits (based on initial lab experiments) for the final API 5, as captured in Table 1. Furthermore, CMAS were made robust by DoE through optimization of the conversion of compound 3 to compound 4 (as described in the companion article). Step 3.2: FMEA-2 for the Identification of Important PP5. The synthesis of API hydrochloride 5 starts with compound 4 as an input and a KSM (Scheme 2); hence, it is important to consider the entire process for conducting an effective FMEA A brief description of the manufacturing process for API hydrochloride 5 is given below: Ethyl acetate and isopropyl alcohol are added to a round-bottom flask. Crude compound 4 is added, and the mixture is stirred for 15 min to achieve complete dissolution. The solution is then cooled to 25 ± 10 °C, and ethyl acetate/HCl is added at the same temperature. The reaction mass is maintained for another 2 h at 25 ± 10 °C. The precipitated API hydrochloride salt is filtered under a nitrogen atmosphere, and the cake is washed with ethyl acetate. The material is then vacuum-dried, unloaded into a vacuum tray drier, and further dried at 47.5 ± 2.5°C until a constant weight is obtained. In order to identify the important input PPS for the DoE studies from the list of all PPS, it was prudent to consider separately each unit operation involved in the manufacturing process for its effect on each CQA using FMEA. Each unit operation became the input for the FMEA, as shown in Table 2. The output of the FMEA was the identification of important input PPS from the list of all the PPS on the basis of the RPN. All of these PPS with the highest RPNs became the input variables for the DoE studies, as shown in Table 3. Step 4. Optimization of the Effects of the Important Input Variables (PP5 and MA^ on CQAs. Step 4.1. Optimization of the Important Input MAS. Since compound 4 and EtOAc/HCl were the KSMs for the reaction, the quality of both KSMs was critical for the reaction. Therefore, acceptance criteria were defined for both KSMs. The acceptance criteria for HC1 and compound 4 were decided on the basis of laboratory "what-if studies as described in Table 2, where maximum tolerable limits were established for every individual impurity. Hence, the important qualities of both as described in Table 2 were taken as CMAS. Step 4.2. Optimization of the Effects of the Important Input PPS on CQAs. The FMEA analysis of the process (FMEA-2; Table 2) revealed three important input PPS on the basis of highest RPN (Table 3); hence, th ese were taken as input variables for the DoE studies for further optimization. The other PPS were kept at predefined levels as captured in Table 4. The output of the DoE is CPPS. It is important to note that each of the three PPS may or may not be included in CPPS. The 23 full factorial design and the results of same are captured in Table 5. It is clear from the results that except for the yield, all Table 5. The 23 full factorial design for optimization of the API reaction conditions S. no. IPA volume cooling temp. (°C) amount of HCl (equiv) yield (%) purity (%) unreacted 4 (%) impurity 6 (%) impurity 7 (%) impurity 8 1 2 5 1 78.37 99.96 0.01 0.02 0 0 2 2 5 1.2 80.54 99.89 0.01 0.01 0.03 0 3 2 35 1.2 74.01 99.98 0 0 0.04 0 4 0.8 5 1 82.72 99.85 0.01 0.05 0 0 5 0.8 35 1 82.72 99.94 0.01 0.01 0.03 0 6 2 35 1 80.54 99.92 0.01 0.01 0.01 0 7 0.8 5 1.2 87.07 99.96 0.01 0.03 0 0 8 0.8 35 1.2 87.07 99.99 0.01 0 0.02 0 9 1.4 20 1.1 87.07 99.97 0.01 0.02 0.01 0 10 1.4 20 1.1 84.90 99.97 0.01 0.02 0.02 0 11 1.4 20 1.1 87.07 99.93 0.01 0.01 0.03 0 12 1.4 20 1.1 84.90 99.95 0.01 0.01 0.02 0 1641 DOI: 10.1021/op500295a Org. Process Res. Dev. 2015,19,1634-1644 Organic Process Research & Development Concept Article Table 6. ANOVA of the 23 full factorial design for yield optimization source sum of squares degrees of freedom mean square F value p value prob > F model 80.27 3 26.76 18.09 0.0011 significant A (IPA volume) 61.28 1 61.28 41.42 0.0004 C (equiv of HCl) 8.69 1 8.69 5.87 0.0459 AC 10.30 1 10.30 6.96 0.0335 curvature 39.65 1 39.65 26.80 0.0013 significant residual 10.36 7 1.48 lack of fit 5.62 4 1.40 0.89 0.5614 not significant pure error 4.74 3 1.58 cor total 130.28 11 Table 7. Additional experiments to augment the 23 full factorial design to RSM point type volumes of IPA addition temp. (°C) equiv of HC1 yield (%) axial 1.4 20 1.2 78.67 axial 2 20 1.1 78.67 axial 0.8 20 1.1 83.04 centre 1.4 20 1.1 84.90 axial 1.4 35 1.1 80.85 axial 1.4 5 1.1 83.04 axial 1.4 20 1 80.85 centre 1.4 20 1.1 85.22 of the CQAs were almost constant (no apparent variation was observed) and well within the specification limits as described in Figure 8. This observation was quite obvious, as these impurities were not formed during the reaction but came from the KSM (compound 4).14 Another process-related lactam impurity, 8, was not observed with the current process. Because of the above observations, only the yield variation (which is also an important CQAa) was analyzed as explained below. The analysis of variance (ANOVA) showed that the yield was a function of the IPA volume and the amount of HC1 (Table 6). It was also evident that the interaction effect of both of the above input PPS was significant. Another important observation was that the curvature was significant. Variation of the addition temperature between 5 and 30 °C did not have any impact on the yield. This indicates that the IPA volume and the amount of HQ were CPPS whereas the temperature was not. Since the curvature was significant (Table 6), the model was augmented to response surface methodology (RSM) with six axial points and two more centre points (Table 7). The ANOVA analysis of the RSM design (Table 8) showed that only the IPA volume plays a significant role in dictating the yield, whereas the amount of HC1 between 0.8—1.4 equiv does not have any significant effect as such, although its interaction with IPA volume is significant. This is also evident by the regression results Table 9. Validation of the model yield (%) run volumes of IPA addition temp. (°c) equiv of HCl actual predicted (95% CI) 1 1.5 10 1.1 83.9 82-85 2 1.5 15 1.1 84.3 82-85 3 1.5 20 1.1 84.7 82-85 4 1.5 15 1.1 83.9 82-85 (eq 3), and the same has been depicted as a contour plot in Figure 9. % yield = -623.97 + 58A + 1240.5C - 61AC - 531C2 (3) Finally, a design space was generated from the contour plot with predefined constraints (0.8—1.4 volumes of IPAand 1 — 1.14 equiv of HC1), as shown in Figure 10, and the model was validated as captured in Table 9. The results of the validation were as expected, and the results were within the 95% confidence interval (Cl). Hence, the volume of IPA and the amount of HC1 became CPPS, which needed to be monitored during commercialization. Step 5. Control Strategy. Finally, a control strategy was planned for all of the CMAS and CPPS. The control strategy for CMAS (compound 4) is already described in Table l,b whereas the control strategies for the other PPS, which were not important, are captured in Table 4. The control strategies for the current two important CPPS are given in Table 10 along with the control strategy for the addition temperature, which was found to be unimportant in a given range. Step 6. FMEA-3: Assessing the Outcome of the Risk Mitigation. The penultimate step of the QbD process is to assess the effect of DoE on the RPN and compare it with the RPN obtained from FMEA-2. For the three CPPS, the RPN decreased significantly (compare the values shown in Table 10 with those in Table 2). Table 8. ANOVA of the percent yield after RSM design source sum of squares degrees of freedom mean square F value p-value prob > F model 131.09 4 32.77 10.05 0.0008 significant A 92.74 1 92.74 28.44 0.0002 C 7.61 1 7.61 2.33 0.1524 AC 35.77 1 35.77 10.97 0.0062 C2 68.05 1 68.05 20.87 0.0006 residual 39.14 12 3.26 lack of fit 31.87 7 4.55 3.13 0.1136 not significant pure error 7.27 5 1.45 1642 DOI: 10.1021/op500295a Org. Process Res. Dev. 2015,19,1634-1644 Organic Process Research & Development Concept Article Table 10. Final CPPs and their control strategies risk assessment 3a S. no. factor acceptable range control strategy O S D RPN 1 IPA volume 0.8—1.6 volumes only the calculated quantity of IPA is to be dispensed using a flow meter with additional calibration on the day storage tank 2 9 2 36 2 addition temperature^ 10-30 °C found to be noncritical from DoE, but the target was set at 20 °C 2 2 2 8 3 amount of HC1 1.04-1.14 equiv assay of EtOAc/HCl solution to be done just before the batch is started to account for any HC1 loss upon storage 2 5 2 20 "O = occurrence, S = severity, D = detectability (not). bAfter DoE, temperature was found not to be a CPP. Step 7. Continuous Improvement: Monitoring both CQAs and CPP5. Since this product is yet to be scaled-up, there are no data to present here. However, there is proper planning at hand in which it has been decided to monitor the CMAS for the API stage for each batch and also to monitor both the CPPS and CQA of the final API using an I-MR control chart. This I-MR control chart is most suitable for the API,8 as any abnormality observed would be recorded and rectified. ■ CONCLUSION This article emphasizes the application of QbD in controlling the CQAs of an API. It is evident from the article that a CQA is dictated by the CPP and CMA, and hence, its identification is critical for any robust process development. This article provides a possible sequence of steps for QbD implementation and illustrates how FMEA could be used for the unbiased selection of important PPs and MAs purely on the basis of the risk assessment of each PP and MA on the CQA. These important PPs can then be used as inputs for the DoE studies for further optimization and help in mitigating the risk associated with them. Additionally, performing FMEA-3 to verify the risk mitigation due to CMAs and CPPs ensures minimum risk to CQAs. Another important consideration for the QbD is the quality of input material from the vendor, and this aspect is considered in the companion article. ■ AUTHOR INFORMATION Corresponding Authors *Telephone: +919701346355. Fax: + 91 08458 279619. E-mail: amrendrakr(2>drreddys.com (A.K.R). *E-mail: sripabba85(2>yahoo.co.in (P.S.). Notes The present article represents the authors' personal views on the subject. The authors declare no competing financial interest. DRL Communication Number IPDO-IPM 00422. ■ ACKNOWLEDGMENTS We thank DRL management for supporting this initiative. ■ ABBREVIATIONS 4A's acceptability, affordability, availability, and accessibility 95% CI confidence interval with an in-built error (a) of 5% ANOVA analysis of variance API active pharmaceutical ingredient CAPA corrective and preventive action CMA critical material attribute CMAj critical material attributes for stage 5 COA certificate of analysis CPP critical process parameter CPPS critical process parameters for stage 5 CQA critical quality attribute DoE design of experiments equiv equivalents FMEA failure mode and effect analysis Hrs hours I-MR individual-moving range KRA key responsibility area KSM key starting material LCL lower control limit LSL lower specification limit MAS material attributes for stage 5 ML mother liquor MVA multivariate analysis NCE new chemical entity NLT not less than NMT not more than OOS out of specification OOT out of trend PAT process analytical tool PP process parameter PPS process parameters for stage 5 QbD quality by design QFD quality function deployment QTPP quality target product profile RPN risk priority number RSM response surface methodology UCL upper control limit USL upper specification limit variance ■ ADDITIONAL NOTES "Yield is a quality parameter as it affects the timely availability of the medicine in the market at a desired price, which is cricial for all stakeholders (i.e., patients, regulators, and the manufacturer). The most important stakeholders are the regulators for the following reasons: The sole responsibility of regulators towards their citizens is to ensure not only acceptable (i.e., good quality) and affordable medicines but also availability of the medicines (no shortages) in their country at all points of time. The responsibility goes even beyond that. The medicines must be easily accessible to patients at their local pharmacies. These four requirements—acceptability, affordability, availability, accessibility (the 4A's)—are a KRA of any regulatory body. If they miss any one of the above 4A's, they will be held accountable by their government for endangering the lives of patients. Also, in eneral, the yield and impurity profile are inversely related. Control of the CMAS of compound 4 is described in the companion article. ■ REFERENCES (l) (a) Musters, J.; Bos, L.; Kellenbach, E. Org. Process Res. Dev. 2013, 17, 87. (b) Cimarosti, Z.; Bravo, F.; Castoldi, D.; Tinazzi, F.; Provera, S.; Perboni, A.; Papini, D.; Westerduin, P. Org. Process Res. Dev. 2010, 14, 1643 DOI: 10.1021/op500295a Org. Process Res. Dev. 2015,19,1634-1644 Organic Process Research & Development Concept Article 805. (c) Bravo, F.; Cimarosti, Z.; Tinazzi, F.; Smith, G. E.; Castoldi, D.; Provera, S.; Westerduin, P. Org. Process Res. Dev. 2010,14, 1162. (2) Mohanty, S.; Roy, A. K.; Kumar, V. K. P.; Reddy, S. G.; Karmakar, A. C. Tetrahedron Lett. 2014, 55, 4585. (b) Mohanty, S.; Roy, A. K.; Kiran, S. P.; Rafael, G. E.; Kumar, V. K P.; Karmakar, A C. Org. Process Res. Dev. 2014, 18, 875. (3) Deshpande, G. R; Roy, A. K; Rao, N. S.; Rao, B. M.; Reddy, J. R. Chromatographia 2011, 73, 639. (4) JuranJ. M. Juran on Quality by Design: The New Steps for Planning Quality Into Goods and Services; Simon and Schuster: New York, 1992. (5) Rodriguez-Perez, J. Quality Risk Management in FDA-Regulated Industry; ASQ_Quality Press: Milwaukee, WI, 2012. (6) (a) ICH Q8 Pharmaceutical Development (R2); U.S. Department of Health and Human Services, Food and Drug Administration, Center for Drug Evaluation and Research (CDER): Rockville, MD, 2009. (b) ICH Q9 Quality Risk Management; U.S. Department of Health and Human Services, Food and Drug Administration, Center for Drug Evaluation and Research (CDER): Rockville, MD, 2006. (c) ICH Q10 Pharmaceutical Quality System; U.S. Department of Health and Human Services, Food and Drug Administration, Center for Drug Evaluation and Research (CDER): Rockville, MD, 2009. (d) Understanding Challenges to Quality by Design; Final deliverable for FDA Understanding Challenges to QbD Project, Dec 18, 2009. (7) Regulators negotiate with the drug product/substance manufacturers on the behalf of their fellow citizens. Hence, they not only have to ensure the acceptability (or quality) and affordability but also the availability of the drug product/substance in their country at all times. Finally, it is desired that the drug product should be accessible to all patients at their local pharmacies. (8) ICH Qll Development and Manufacture of Drug Substances; U.S. Department of Health and Human Services, Food and Drug Administration, Center for Drug Evaluation and Research (CDER): Rockville, MD, 2012. (9) Mukundam, K; Varma, D. R N.; Deshpande, G. R; Dahanukar, V. H; Roy, A K. Org. Process Res. Dev. 2013,17, 1002. (10) Kubiak, T. M.; Benbow, D. W. The Certified Six Sigma Black Belt Handbook, 2nd ed.; ASQ,Quality Press: Milwaukee, WI, 2009. (11) (a) Lobben, P. C; Barlow, E.; Bergum, J. S.; Braem, A; Chang, S. Y.; Gibson, F.; Kopp, N.; Lai, C; LaPorte, T. L.; Leahy, D. K; Miislehiddinoglu, J.; Quiroz, F.; Skliar, D.; Spangler, L.; Srivastava, S.; Wasser, D.; Wasylyk, J.; Wethman, R; Xu, Z. Org. Process Res. Dev. 2014, DOI: 10.1021/op500126u. (b) Zhou, G.; Moment, A; Cuff, J.; Schafer, W.; Orella, C; Sirota, E.; Gong, X.; Welch, C. Org. Process Res. Dev. 2014, DOI: 10.1021/op5000978. (12) Deshpande, A A.; Ramya, A; Vishweshwar, V.; Deshpande, G. R.; Roy, A K Org. Process Res. Dev. 2014,18, 1614-1621. (13) As stated earlier, the yield is a quality parameter, and another point to be noted here is the quantity of IPA to be charged, which has a very narrow range (1—1.5 volumes). A high volume of IPA results in yield loss, where as a low volume of IPA results in unreacted free base (compound 4 or impurity 4) because of the thick slurry nature of the reaction mass. Low volumes also created problems during the transfer of the reaction mass from the reactor to the filter. A variation was observed when IPA was dispensed from the day tank. Hence, a flow meter was proposed to measure the volume accurately. This explains the high severity and high detectability. (14) Many times it happens that DoE fails to provide a solution to control an impurity during a reaction (i.e., there is no effect of PPs on the impurity). In such situations there could be two reasons for DoE failure. The first is that one is working with the wrong variables, and the second is that the impurities come from earlier stages. 1644 DOI: 10.1021/op500295a Org. Process Res. Dev. 2015,19,1634-1644