Using Patient Data in Amyloidosis to Understand Complex Diagnosis Pathways and Treatment Patterns

Amyloidosis is a rare disease that occurs when a substance called amyloid builds up in the body’s organs. Amyloid is an abnormal protein that is produced in bone marrow and can be deposited in any tissue or organ, affecting their normal function. The disease consists of many different sub-types and the type of protein that is misfolded along with the organ or tissue in which the misfolded proteins are deposited determines the clinical manifestations of amyloidosis.

Without treatment, amyloid fibrils accumulate and lead to organ impairment, failure, and ultimately death. The rarity of the disease and the multi system presentation of the disease are believed to lead to a large number of late or undiagnosed patients. In subtypes such as AL amyloidosis, urgent diagnosis and treatment is essential to improve patient outcomes. Therefore, finding new ways to help improve detection and diagnosis will greatly improve patient's outcomes.

Parties involved:

Each party in the collaboration has a different role during the research:

1) The National Amyloidosis Centre (NAC) held at Royal Free London NHS Foundation Trust will support IQVIA Ltd ‘s ethical approvals activities; provide expert clinical insight on the research findings and support on NAC dataset de-identification in addition to supporting the publication of research findings. Royal Free Hospital provides services to UCL staff.

IQVIA Ltd were added as a joint Data Controller to the previous agreement as IQVIA simplified the number of trading legal entities it had in the UK by transferring the business and assets from IQVIA Solutions UK Ltd into IQVIA Ltd.

IQVIA Ltd is the legacy Quintiles part of the IQVIA business that runs the clinical trials.

Where the term IQVIA is used in this agreement, it is referring to any or both of IQVIA Technology Services Ltd and IQVIA Ltd.

2) IQVIA will conduct ethical approval activities;

conduct the transformation and data processing of de-identified data into analysable format, and perform the analysis described in this document. IQVIA has significant experience with HES data, other retrospective databases, outcomes research expertise and advanced machine learning capabilities for predictive algorithm development. IQVIA will provide infrastructure and technical support to allow the hosting of de- identified HES and de-identified NAC data

3) GSK used to be the industry partner who previously, but no longer, sponsored this research. The research will now be conducted and funded by IQVIA and the Royal Free Hospital.

The data linkage was carried out under a previous version of this agreement. The continued processing and analysis of such linked data is necessary for the purposes of the legitimate interests pursued by the controller or by a third party, except where such interests are overridden by the interests or fundamental rights and freedoms of the data subject which require protection of personal data, in particular where the data subject is a child (covered by Article 6 (1)(F) of the GDPR).

IQVIA Limited [and its affiliates] provide information and technology services to healthcare. IQVIA Limited produces a longitudinal research database and requires access to health data to serve its purpose of serving the health care industry and populating the longitudinal research database.

The Legal basis for processing of personal data relating to patient health is under Article 9(2)(j)of the GDPR - Processing is necessary for archiving purposes in the public interest, scientific or historical research purposes or statistical purposes in accordance with Article 89(1) based on Union or Member State law which shall be proportionate to the aim pursued, respect the essence of the right to data protection and provide for suitable and specific measures to safeguard the fundamental rights and the interests of the data subject. The data subjects’ interests and fundamental rights are protected through appropriate minimisation of fields; protection of the data in a secure environment, and guaranteeing secure destruction at any stage at the request of NHS Digital or after a defined period on completion of the project.

The NAC (as England's only amyloidosis diagnostic and treatment centre) benefits from the research by furthering an understanding of patients's journeys outside the NAC, supporting improvements in detection of amyloidosis and supporting better referral to the NAC by educating and sharing learning with other hospitals who may see undiagnosed amyloidosis patients.

Data access will be restricted to substantive employees of IQVIA Ltd. and IQVIA Technology Services Limited in their capacity as data processors; they have this role due to their significant experience in patient pathway analytics, generation of predictive algorithms and working with HES data.

The analysis will be conducted by IQVIA, This analysis will be conducted based on the pre-defined contractually agreed protocol. IQVIA has capabilities in the area of predictive analytics and pathway analysis and has developed bespoke sets of methodology which is expert-driven from a group of employees with a strong academic and data science background. Critically, they have focused on model interpretation which is not a priority in the machine learning field but in healthcare the interpretation is imperative.

The patient selection criteria have been based on patients who attended NAC and those who have visited specialists which are frequented by patients with amyloidosis. This process has been developed and chosen by IQVIA Ltd . The dissemination of findings has been pre-agreed and outlined in the outputs section.

The data retention period has been agreed in this data sharing agreement.

If IQVIA requires more time for the analysis they will request an extension to this agreement.

The aims of the research are to:

1) Understand the amyloidosis patient's diagnostic pathway and outcomes. This includes the implications of going through different routes to diagnosis, which can be used to develop materials which can help educate physicians on how to diagnose patients earlier;

2) Identify barriers in the patient pathways to receiving diagnosis

3) Understand current coding in HES for different subtypes of amyloidosis, which can be used to support applications to change current ICD-10 coding practices in the UK and therefore enable capturing of more clinically accurate patient information nationally, which can support future research efforts in this understudied condition;

4) Develop a predictive algorithm which would be able to flag patients with a high probability of having amyloidosis (and subtypes) from their data fingerprint, which will support finding undiagnosed patients.

In order to achieve the goals listed above, IQVIA has successfully linked the HES data to the National Amyloidosis Centre (NAC) dataset under a previous version of this agreement. This allowed IQVIA to create a combined

dataset for research to better understand and improve the detection and treatment of amyloidosis.

Dissemination of results will be guided by the clinical interpretation group and if an effective predictive algorithm is produced, then efforts will be made to implement this in an appropriate manner given its capabilities. The interpretation group is made up of personnel from Royal Free Hospital and IQVIA, who conduct regular meetings to discuss study related matters (e.g. study outputs/design).

Linking HES data with the NAC dataset has utilised the confirmed and sub-typed NAC patient diagnoses present in the NAC dataset, where the patient amyloid classification has been confirmed by world leading clinical experts. This has allowed IQVIA to identify patients with confirmed amyloidosis (and subtypes of amyloidosis) within the HES data for investigation and analysis with high certainty.

Current ICD-10 coding (the International classification system for coding of disease types, maintained by the World Health Organisation) does not have a specific code for amyloidosis subtypes (i.e. Familial Amyloid Cardiomyopathy (FAC), Familial Amyloid Polyneuropathy (FAP), Amyloid light-chain (AL) amyloidosis), with multiple different subtypes coded under the same ICD-10 code. These subtypes have dramatically different outcomes and patient pathways and thus being able to differentiate the patients is key to the research.

The data requested has been filtered to:

1) Cohort A: Patients with confirmed amyloidosis, which consists of:

a) Consented Participants in the NAC database. Identifiers have been sent to NHS Digital in order to link study ID only to the HES data

b) Patients who have an amyloidosis diagnosis code who have not attended the Royal Free (sourced from the HES database)

2) Cohort B: Patients with unconfirmed amyloidosis, which consists of the patients who visit specialities often visited by patients with an amyloidosis diagnosis (based on the presence of E85 ICD-10). Data will be restricted to only include patients holding 1 or more of 22 specialities of which have been visited at some point in time by the vast majority (>97%) of amyloidosis patients. This is required due to the rarity of the disease and research has found that patients can have a 7+ year diagnosis process due to the variety and complexity of symptoms. This will be sourced from the HES database.

IQVIA Ltd selected a broad range of variables as when developing a predictive algorithm, the factors which may act as a data fingerprint are unclear until the process has started, removing particular variables thus can impact the power of the predictive algorithm, thus potentially detect patients with a high risk of having undiagnosed amyloidosis much later than if a full suite of variables was available.

The joint data controllers are requesting to continue to hold ~15-year historical extract of data for the amyloidosis project for both participants in the NAC database and patients who meet the criteria in the extract. The reason being that amyloidosis patient populations (especially at subtype level) are very small and the diagnosis pathway is a long multi-year often complex process. This requires HES data for a longer time period in order to capture sufficient patients for the analysis.

In amyloidosis physicians often do not initially attribute symptoms present to the rare disease in question. This means that patients are often misdiagnosed and seen by multiple physicians before an accurate diagnosis is made. In some cases, patients will have a diagnosis process that takes years due to the variety and complexity of symptoms e.g. in SSA it has been shown that it can be 5.4 to 4.4 years from onset to diagnosis (Nakagawa et al., 2016). >15 years of data will facilitate more robust and insightful analysis of these types of patient groups and provide a better grounding for potential earlier diagnosis interventions in future.

Due to the complicated disease area and the need to create a sophisticated algorithm that has the potential to perform well in the live clinical environment, a large sample of data is required. Below is an overview of the reasons for the selection criteria:

1) Disease characteristics:

The aim was to create a predictive algorithm for multiple different amyloidosis subtypes [Amyloidosis Light-chain (AL), Familial amyloidotic polyneuropathies (FAP), Familial amyloid cardiomyopathy (FAC), Senile Systemic Amyloidosis (SSA)].

Between these subtypes and even within these subtypes, patients can exhibit large differences in clinical presentation.

Even in the more defined Familial amyloid cardiomyopathy (FAC), and/or Amyloid transthyretin amyloid cardiomyopathy (ATTR) subtypes, patients can exhibit GI and autonomic nervous system involvement in addition to the cardiac symptoms presented. Patients with FAP usually present between the ages of 20 and 40 whereas patients with SSA often present past the age of 70. This means that a large range of specialities, symptoms, procedures and demographics need to be assessed when generating algorithms and defining the cohorts, as each subtype will require its own comparison cohort, selected from cohort B.

2) Refining the comparison cohort (cohort B) based on clinical characteristics of the particular subtype:

In order to select the most appropriate cohort of patients to act as a comparison group to the confirmed amyloidosis patients (cohort A), IQVIA will analyse the patient data. This is a data-driven exploratory approach that

will allow the joint data controllers to select the most appropriate patients for cohort B. As noted previously amyloidosis patients are often misdiagnosed as other conditions due to the rarity of the disease and huge range of clinical manifestations they can present with. The aim is to identify a cohort of patients which did not have a confirmed diagnosis but share very similar clinical features; for example: have contaminant diagnoses, visit the same specialists etc.

This will allow the algorithm to be developed on a comparison group as close to the real cases physicians experience in clinical practice as possible and thus ensure any algorithm developed is as robust as possible. For example, IQVIA created an algorithm to identify a rare disease population (Idiopathic Pulmonary Fibrosis), which manifests as a lung condition commonly misdiagnosed as asthma or COPD (Chronic Obstructive Pulmonary Disease). To focus the algorithm on the clinical challenge IQVIA developed the algorithm to distinguish between IPF patients and those with COPD/Asthma.

Amyloidosis is a significantly more complex disease than the previous example and requires a deep dive into the data to align the patient cohorts

3) Refining the cohorts (both A and B) based on availability of appropriate length of historic data:

The size of a cohort is limited not only by the number of patients with given diagnosis, but also the need to have available a sufficient time period both prior to the diagnosis (to observe baseline characteristics) as well as after the event (to observe relevant outcomes) for analysis. For example, a recent project in Fabry Disease, one focus of analysis was to understand the diagnostic pathway, in order to identify any predictive signals/ markers which would allow earlier diagnosis of Fabry disease and thus slowing progression of the disease by allowing earlier treatment. Fabry Disease is an inherited disorder that results from the build-up of a particular type of fat, called globotriaosylceramide, in the body's cells. It usually begins in childhood and affects many parts of the body, while can potentially be life-threatening due to progressive kidney damage, heart attack, and stroke. The study by IQVIA identified 665 patients with suspected Fabry disease; of those patients only 90 patients had 3+ years of historical data available to allow analysis of the lead up to patient diagnosis (which was much shorter than desirable given the often 20-year symptom onset in this condition). This patient cohort size prevented IQVIA from having sufficient numbers to conduct robust predictive analytics on the data to find signals/ markers of disease.

Through the years of managing and diagnosing amyloidosis patients, the Honorary Consultant Nephrologist at the NAC, has noticed that > 50% of cases of patients with the FAC subtype of amyloidosis have prior carpel tunnel syndrome/ decompression occurrence in patients up to 10 years prior to diagnosis. It is IQVIA’s hypothesis that this in conjunction with other attributes may act as a predictive marker of early FAC disease. Due to the length of the time frame, IQVIA Ltd. has requested more than 15 years of HES data.

4) Bringing the algorithm to clinical practice:

If the algorithm were to be implemented in a live clinical practice setting, the algorithm can only run on patients who fit inclusion and exclusion criteria used to pull HES data. Therefore, the smaller the patient sample that is requested, the more limited the real-world sample will be and assessed for the risk of the disease. For example, if a sample of HES data is requested and made up solely of male patients, over 40 years old, this would mean that the model could not be expected to produce robust predictions for any female patients or patients under the age of 40, thus limiting the potential benefits of the outputs.

References:

Nakagawa. M et al, Carpal tunnel syndrome: a common initial symptom of systemic wild-type ATTR (ATTRwt) amyloidosis,

Amyloid. 2016;23(1):58-63. doi: 10.3109/13506129.2015.1135792. Epub 2016 Feb 8.

All organisations party to this agreement must comply with the Data Sharing Framework Contract requirements, including those regarding the use (and purposes of that use) by Personnel(as defined within the Data Sharing Framework Contract ie: employees, agents and contractors of the Data Recipient who may have access to that data)

There will be no data linkage undertaken with NHS Digital data provided under this agreement that is not already noted in the agreement.

Data will only be accessed and processed by substantive employees of IQVIA Technology Services Ltd and IQVIA Ltd and will not be accessed or processed by any other third parties not mentioned in this agreement.

Since the initial approval of this Agreement in April 2017, the analyses completed have addressed the primary objective of the research; to describe the diagnostic and treatment pathways for different amyloidosis subtypes. Transthyretin amyloidosis is a slowly progressive condition characterized by the build-up of abnormal deposits of a protein called amyloid (amyloidosis) in the body's organs and tissues. These protein deposits most frequently occur in the peripheral nervous system, which is made up of nerves connecting the brain and spinal cord to muscles and sensory cells that detect sensations such as touch, pain, heat, and sound. Protein deposits in these nerves result in a loss of sensation in the extremities (peripheral neuropathy). The autonomic nervous system, which controls involuntary body functions such as blood pressure, heart rate, and digestion, may also be affected by amyloidosis. In some cases, the brain and spinal cord (central nervous system) are affected. Other areas of amyloidosis include the heart, kidneys, eyes, and gastrointestinal tract, This analysis has been completed for multiple subtypes of Amyloidosis:

1) Cardiac transthyretin amyloidosis (ATTR-CM)

2) Neuropathic transthyretin amyloidosis (ATTR-PN)

3) Wild type transthyretin amyloidosis (wtATTR)

4) Systemic light chain amyloidosis (AL)

Findings in the ATTR-CM sub-type have been written up in a manuscript entitled Natural history, quality of life and

outcomes in cardiac ATTR amyloidosis, that has been published to the Circulation in July 2 2019

(https://www.ahajournals.org/doi/full/10.1161/CIRCULATIONAHA.118.038169 ). Findings from the analysis in the remaining two subtypes, ATTR-PN and AL, will be developed into publications to submit to peer-reviewed journals, as well as further findings from the ATTR-CM subtype.

Outputs will contain only aggregate-level data with small numbers suppressed in line with the HES analysis guide. Outputs are intended to improve the currently limited understanding of the patient journey and diagnostic pathway in amyloidosis, which could realise benefits such as a more efficient care pathway, earlier diagnosis and treatment (this is described in more detail in the expected measurable benefits section). Outputs are for peer review publications and will not be used for commercial / sales and marketing purposes.

The secondary objective of the research was to describe predictive patient characteristics to support:

1. The diagnosis of amyloidosis patients earlier in the patient pathway than they would otherwise be diagnosed

2. The flagging of high-risk patients for diagnostic tests, who otherwise may go un-diagnosed

This objective has yet to be addressed due to limited capacity from academic collaborators and discontinuation of sponsorship from industry collaborators, GlaxoSmithKline. IQVIA notes in this extension agreement that GlaxoSmithKline is no longer funding this research , and that further research will be jointly funded by IQVIA and The Royal Free Hospital.

The purpose of this extension is to secure continued access to the research dataset to:

1. Respond to any further comments from reviewers on the manuscript submitted to Circulation

2. Publish further findings from the ATTR-CM subtype beyond what is included in the first Circulation manuscript

3. Write up findings from the analysis of treatment patterns, treatment toxicity and outcomes in AL patients who received chemotherapy (due for submission September 2019)

4. Write up the findings from the analysis of diagnostic and treatment pathways in other subtypes of the disease into a manuscript for submission to a peer-reviewed journal

5. Address the secondary objective of the research: to describe predictive patient characteristics to support the flagging of patients earlier in their diagnostic pathway, or the flagging of patients who have not yet been diagnosed via the development of a predictive algorithm IQVIA expects to produce the following analyses:

1) Investigate the predictive patient characteristics within the data environment to understand if IQVIA can support the flagging of patients earlier in their diagnostic pathway or flag patients who have not yet been diagnosed via the

development of a predictive algorithm expected to be completed 12-18 months after HES data has been provided .

The target dissemination plan is as follows:

2) The applicant will submit the findings of the research to a peer review journal e.g. Rheumatology

3) The applicant will submit and present on findings at a relevant amyloidosis conference e.g. 2020 International

Symposium on Amyloidosis, in order to further the knowledge of other specialist physicians

4) Published results will be shared with the UKAAG patient advocacy group

5) Furthermore the abstracts and links to publications will be hosted on IQVIAs online bibliography which is publicly available

6) Results will also be shared with other parties where appropriate e.g. Sharing results with other NHS trusts who also manage amyloidosis patients or sharing with international centres which also diagnose and manage amyloidosis patients

The output of the algorithm generation is currently uncertain. However, any implementation would need to be conducted by or with NHS bodies, because IQVIA is working with pseudonymised data and will not seek to re-identify patients at any stage. The nature of any implementation would need to be driven by the predictive sensitivity and specificity of the algorithm. In other words, the false positive and false negative detection rate. Implementing an algorithm with a high false positive rate would lead to many people tested with very few identified, conversely if the algorithm has a high false negative rate, it will likely miss many patients who should be tested for the disease. The health economics of the algorithm and any associated intervention would need to be carefully assessed prior to any implementation. Prior to any algorithm playing a role in supporting clinical practice / being implemented it will require peer review publication and broad acceptance before any uptake could be successful.

For an algorithm with weaker predictive potential, IQVIA envisages the generation of publications in peer reviewed journals and generation of medical educational materials to use with clinical specialities who are potentially exposed to amyloidosis patients. The literature will document the methodology used and the risk factors which would help to identify amyloidosis patients earlier. These will potentially be presented at symposiums or other forums, depending on the findings.

If an algorithm with high predictive potential is generated, it could be used to create a clinical support tool for physicians to help diagnose patients, allowing the summarisation of large quantities of data in a more manageable format. This tool could support physicians by providing a risk score which they can interpret themselves to support clinician decisions.

The algorithm would be free of charge and openly available. Access methods would be dependent on the strength of the algorithm but may include presentation at seminars, publications on risk factors or a clinical support tool provided directly to physicians (subject to any relevant approvals) If no information of merit is found, the methodology utilised in the research would be documented and submitted to a peer reviewed journal. This would allow other researchers to benefit from the research efforts. In addition, the methodology would be shared via IQVIA's online bibliography and which is publicly available. In all summaries, any data used would be aggregated with small numbers suppressed in line with the HES Analysis Guide.

No organisation on the clinical interpretation group will have the ability to suppress the dissemination of findings or outputs from this work. None of these outputs are linked to PhD studies.

No further yielded benefits have been realised as the findings have been published. The data is required to answer any potential enquiries relating to the publications.

Based on the data provided for the previous agreement, IQVIA has developed a full set of analysis regarding the NAC patient pathway for Amyloidosis which have been discussed with both the NAC (the academic collaborators) and GSK (an industry partner who previously, but no longer, sponsored this research). As a result of these discussions, there is a considerably better understanding of the Amyloidosis patient pathway, including per subtype.

For example, for ATTR-CM patients treated at the NAC:

- There is a substantial delay in diagnosis following onset of symptoms, with patients using hospital services (either as an inpatient, outpatient or Accident & Emergency visit) a mean of 19.9 times during the 3 years before diagnosis; diagnosis of the wild-type form of ATTR-CM was delayed more than 4 years after onset of cardiac

symptoms in 42% of cases

- Hereditary cardiac amyloidosis patients with a certain genetic mutation (V122I) were more impaired functionally and had worse measures of cardiac disease at the time of diagnosis, and poorer survival compared to the other sub-groups (such as patients carrying a different genetic mutation, and patients with wild-type form of the disease)

- Analysis of the diagnostic and treatment pathways for different amyloidosis subtypes has also been completed after HES data was provided, leading to improved knowledge and patient benefits.

All parties now have a better understanding of treatments that occur outside of the NAC and patient outcomes. A set of papers from the outputs of the analysis have been already published as mentioned above or will be developed into publication and published in research journals, with one manuscript, entitled Natural history, quality of life and outcomes in cardiac ATTR amyloidosis, already published in the journal Circulation. Through publication, the results have been shared across the scientific community, increasing the body of knowledge on Amyloidosis disease and benefiting Amyloidosis patients with potential future treatment innovations.

Amyloidosis is a rare disease that occurs when a substance called amyloid builds up in the body’s organs. Amyloid is an abnormal protein that is produced in bone marrow and can be deposited in any tissue or organ, affecting their normal function. The disease consists of many different sub-types and the type of protein that is misfolded along with the organ or tissue in which the misfolded proteins are deposited determines the clinical manifestations of amyloidosis.

Without treatment, amyloid fibrils accumulate and lead to organ impairment, failure, and ultimately death. The rarity of the disease and the multi system presentation of the disease are believed to lead to a large number of late or undiagnosed patients. In subtypes such as AL amyloidosis, urgent diagnosis and treatment is essential to improve patient outcomes. Therefore, finding new ways to help improve detection and diagnosis will greatly improve patient's outcomes.

Parties involved:

Each party in the collaboration has a different role during the research:

1) The National Amyloidosis Centre (NAC) held at Royal Free London NHS Foundation Trust will support IQVIA Solutions UK Ltd ‘s ethical approvals activities; provide expert clinical insight on the research findings and support on NAC dataset de-identification in addition to supporting the publication of research findings. Royal Free Hospital provides services to UCL staff.

IQVIA Ltd has been added as a joint Data Controller to this agreement as IQVIA are simplifying the number of trading legal entities it has in the UK by transferring the business and assets from IQVIA Solutions UK Ltd into IQVIA Ltd.

IQVIA Ltd is the legacy Quintiles part of the IQVIA business that runs the clinical trials.

Where the term IQVIA is used in this agreement, it is referring to any or all of IQVIA Technology Services Ltd., IQVIA Solutions UK Ltd, and with effect from the date of this agreement, IQVIA Ltd.

2) IQVIA will conduct ethical approval activities;

conduct the transformation and data processing of de-identified data into analysable format, and perform the analysis described in this document. IQVIA has significant experience with HES data, other retrospective databases, outcomes research expertise and advanced machine learning capabilities for predictive algorithm development. IQVIA will provide infrastructure and technical support to allow the hosting of de- identified HES and de-identified NAC data

3) GSK used to be the industry partner who previously, but no longer, sponsored this research. The research will now be conducted and funded by IQVIA and the Royal Free Hospital.

The data linkage was carried out under a previous version of this agreement. The continued processing and analysis of such linked data is necessary for the purposes of the legitimate interests pursued by the controller or by a third party, except where such interests are overridden by the interests or fundamental rights and freedoms of the data subject which require protection of personal data, in particular where the data subject is a child (covered by Article 6 (1)(F) of the GDPR).

IQVIA Solutions UK Limited [and its affiliates] provide information and technology services to healthcare. IQVIA Solutions UK Limited produces a longitudinal research database and requires access to health data to serve its purpose of serving the health care industry and populating the longitudinal research database.

The Legal basis for processing of personal data relating to patient health is under Article 9(2)(j)of the GDPR - Processing is necessary for archiving purposes in the public interest, scientific or historical research purposes or statistical purposes in accordance with Article 89(1) based on Union or Member State law which shall be proportionate to the aim pursued, respect the essence of the right to data protection and provide for suitable and specific measures to safeguard the fundamental rights and the interests of the data subject. The data subjects’ interests and fundamental rights are protected through appropriate minimisation of fields; protection of the data in a secure environment, and guaranteeing secure destruction at any stage at the request of NHS Digital or after a defined period on completion of the project.

The NAC (as England's only amyloidosis diagnostic and treatment centre) benefits from the research by furthering an understanding of patients's journeys outside the NAC, supporting improvements in detection of amyloidosis and supporting better referral to the NAC by educating and sharing learning with other hospitals who may see undiagnosed amyloidosis patients.

Data access will be restricted to substantive employees of IQVIA Solutions UK Ltd, IQVIA Ltd. and IQVIA Technology Services Limited in their capacity as data processors; they have this role due to their significant experience in patient pathway analytics, generation of predictive algorithms and working with HES data.

The analysis will be conducted by IQVIA, This analysis will be conducted based on the pre-defined contractually agreed protocol. IQVIA has capabilities in the area of predictive analytics and pathway analysis and has developed bespoke sets of methodology which is expert-driven from a group of employees with a strong academic and data science background. Critically, they have focused on model interpretation which is not a priority in the machine learning field but in healthcare the interpretation is imperative.

The patient selection criteria have been based on patients who attended NAC and those who have visited specialists which are frequented by patients with amyloidosis. This process has been developed and chosen by IQVIA Solutions UK Ltd . The dissemination of findings has been pre-agreed and outlined in the outputs section.

The data retention period has been agreed in this data sharing agreement.

If IQVIA requires more time for the analysis they will request an extension to this agreement.

The aims of the research are to:

1) Understand the amyloidosis patient's diagnostic pathway and outcomes. This includes the implications of going through different routes to diagnosis, which can be used to develop materials which can help educate physicians on how to diagnose patients earlier;

2) Identify barriers in the patient pathways to receiving diagnosis

3) Understand current coding in HES for different subtypes of amyloidosis, which can be used to support applications to change current ICD-10 coding practices in the UK and therefore enable capturing of more clinically accurate patient information nationally, which can support future research efforts in this understudied condition;

4) Develop a predictive algorithm which would be able to flag patients with a high probability of having amyloidosis (and subtypes) from their data fingerprint, which will support finding undiagnosed patients.

In order to achieve the goals listed above, IQVIA has successfully linked the HES data to the National Amyloidosis Centre (NAC) dataset under a previous version of this agreement. This allowed IQVIA to create a combined

dataset for research to better understand and improve the detection and treatment of amyloidosis.

Dissemination of results will be guided by the clinical interpretation group and if an effective predictive algorithm is produced, then efforts will be made to implement this in an appropriate manner given its capabilities. The interpretation group is made up of personnel from Royal Free Hospital and IQVIA, who conduct regular meetings to discuss study related matters (e.g. study outputs/design).

Linking HES data with the NAC dataset has utilised the confirmed and sub-typed NAC patient diagnoses present in the NAC dataset, where the patient amyloid classification has been confirmed by world leading clinical experts. This has allowed IQVIA to identify patients with confirmed amyloidosis (and subtypes of amyloidosis) within the HES data for investigation and analysis with high certainty.

Current ICD-10 coding (the International classification system for coding of disease types, maintained by the World Health Organisation) does not have a specific code for amyloidosis subtypes (i.e. Familial Amyloid Cardiomyopathy (FAC), Familial Amyloid Polyneuropathy (FAP), Amyloid light-chain (AL) amyloidosis), with multiple different subtypes coded under the same ICD-10 code. These subtypes have dramatically different outcomes and patient pathways and thus being able to differentiate the patients is key to the research.

The data requested has been filtered to:

1) Cohort A: Patients with confirmed amyloidosis, which consists of:

a) Consented Participants in the NAC database. Identifiers have been sent to NHS Digital in order to link study ID only to the HES data

b) Patients who have an amyloidosis diagnosis code who have not attended the Royal Free (sourced from the HES database)

2) Cohort B: Patients with unconfirmed amyloidosis, which consists of the patients who visit specialities often visited by patients with an amyloidosis diagnosis (based on the presence of E85 ICD-10). Data will be restricted to only include patients holding 1 or more of 22 specialities of which have been visited at some point in time by the vast majority (>97%) of amyloidosis patients. This is required due to the rarity of the disease and research has found that patients can have a 7+ year diagnosis process due to the variety and complexity of symptoms. This will be sourced from the HES database.

IQVIA Solutions UK Ltd. has selected a broad range of variables as when developing a predictive algorithm, the factors which may act as a data fingerprint are unclear until the process has started, removing particular variables thus can impact the power of the predictive algorithm, thus potentially detect patients with a high risk of having undiagnosed amyloidosis much later than if a full suite of variables was available.

The joint data controllers are requesting to continue to hold ~15-year historical extract of data for the amyloidosis project for both participants in the NAC database and patients who meet the criteria in the extract. The reason being that amyloidosis patient populations (especially at subtype level) are very small and the diagnosis pathway is a long multi-year often complex process. This requires HES data for a longer time period in order to capture sufficient patients for the analysis.

In amyloidosis physicians often do not initially attribute symptoms present to the rare disease in question. This means that patients are often misdiagnosed and seen by multiple physicians before an accurate diagnosis is made. In some cases, patients will have a diagnosis process that takes years due to the variety and complexity of symptoms e.g. in SSA it has been shown that it can be 5.4 to 4.4 years from onset to diagnosis (Nakagawa et al., 2016). >15 years of data will facilitate more robust and insightful analysis of these types of patient groups and provide a better grounding for potential earlier diagnosis interventions in future.

Due to the complicated disease area and the need to create a sophisticated algorithm that has the potential to perform well in the live clinical environment, a large sample of data is required. Below is an overview of the reasons for the selection criteria:

1) Disease characteristics:

The aim was to create a predictive algorithm for multiple different amyloidosis subtypes [Amyloidosis Light-chain (AL), Familial amyloidotic polyneuropathies (FAP), Familial amyloid cardiomyopathy (FAC), Senile Systemic Amyloidosis (SSA)].

Between these subtypes and even within these subtypes, patients can exhibit large differences in clinical presentation.

Even in the more defined Familial amyloid cardiomyopathy (FAC), and/or Amyloid transthyretin amyloid cardiomyopathy (ATTR) subtypes, patients can exhibit GI and autonomic nervous system involvement in addition to the cardiac symptoms presented. Patients with FAP usually present between the ages of 20 and 40 whereas patients with SSA often present past the age of 70. This means that a large range of specialities, symptoms, procedures and demographics need to be assessed when generating algorithms and defining the cohorts, as each subtype will require its own comparison cohort, selected from cohort B.

2) Refining the comparison cohort (cohort B) based on clinical characteristics of the particular subtype:

In order to select the most appropriate cohort of patients to act as a comparison group to the confirmed amyloidosis patients (cohort A), IQVIA will analyse the patient data. This is a data-driven exploratory approach that

will allow the joint data controllers to select the most appropriate patients for cohort B. As noted previously amyloidosis patients are often misdiagnosed as other conditions due to the rarity of the disease and huge range of clinical manifestations they can present with. The aim is to identify a cohort of patients which did not have a confirmed diagnosis but share very similar clinical features; for example: have contaminant diagnoses, visit the same specialists etc.

This will allow the algorithm to be developed on a comparison group as close to the real cases physicians experience in clinical practice as possible and thus ensure any algorithm developed is as robust as possible. For example, IQVIA created an algorithm to identify a rare disease population (Idiopathic Pulmonary Fibrosis), which manifests as a lung condition commonly misdiagnosed as asthma or COPD (Chronic Obstructive Pulmonary Disease). To focus the algorithm on the clinical challenge IQVIA developed the algorithm to distinguish between IPF patients and those with COPD/Asthma.

Amyloidosis is a significantly more complex disease than the previous example and requires a deep dive into the data to align the patient cohorts

3) Refining the cohorts (both A and B) based on availability of appropriate length of historic data:

The size of a cohort is limited not only by the number of patients with given diagnosis, but also the need to have available a sufficient time period both prior to the diagnosis (to observe baseline characteristics) as well as after the event (to observe relevant outcomes) for analysis. For example, a recent project in Fabry Disease, one focus of analysis was to understand the diagnostic pathway, in order to identify any predictive signals/ markers which would allow earlier diagnosis of Fabry disease and thus slowing progression of the disease by allowing earlier treatment. Fabry Disease is an inherited disorder that results from the build-up of a particular type of fat, called globotriaosylceramide, in the body's cells. It usually begins in childhood and affects many parts of the body, while can potentially be life-threatening due to progressive kidney damage, heart attack, and stroke. The study by IQVIA identified 665 patients with suspected Fabry disease; of those patients only 90 patients had 3+ years of historical data available to allow analysis of the lead up to patient diagnosis (which was much shorter than desirable given the often 20-year symptom onset in this condition). This patient cohort size prevented IQVIA from having sufficient numbers to conduct robust predictive analytics on the data to find signals/ markers of disease.

Through the years of managing and diagnosing amyloidosis patients, the Honorary Consultant Nephrologist at the NAC, has noticed that > 50% of cases of patients with the FAC subtype of amyloidosis have prior carpel tunnel syndrome/ decompression occurrence in patients up to 10 years prior to diagnosis. It is IQVIA’s hypothesis that this in conjunction with other attributes may act as a predictive marker of early FAC disease. Due to the length of the time frame, IQVIA Solutions UK Ltd. has requested more than 15 years of HES data.

4) Bringing the algorithm to clinical practice:

If the algorithm were to be implemented in a live clinical practice setting, the algorithm can only run on patients who fit inclusion and exclusion criteria used to pull HES data. Therefore, the smaller the patient sample that is requested, the more limited the real-world sample will be and assessed for the risk of the disease. For example, if a sample of HES data is requested and made up solely of male patients, over 40 years old, this would mean that the model could not be expected to produce robust predictions for any female patients or patients under the age of 40, thus limiting the potential benefits of the outputs.

References:

Nakagawa. M et al, Carpal tunnel syndrome: a common initial symptom of systemic wild-type ATTR (ATTRwt) amyloidosis,

Amyloid. 2016;23(1):58-63. doi: 10.3109/13506129.2015.1135792. Epub 2016 Feb 8.

All organisations party to this agreement must comply with the Data Sharing Framework Contract requirements, including those regarding the use (and purposes of that use) by Personnel(as defined within the Data Sharing Framework Contract ie: employees, agents and contractors of the Data Recipient who may have access to that data)

There will be no data linkage undertaken with NHS Digital data provided under this agreement that is not already noted in the agreement.

Data will only be accessed and processed by substantive employees of IQVIA Solutions UK Limited, IQVIA Technology Services Ltd and IQVIA Ltd and will not be accessed or processed by any other third parties not mentioned in this agreement.

Since the initial approval of this Agreement in April 2017, the analyses completed have addressed the primary objective of the research; to describe the diagnostic and treatment pathways for different amyloidosis subtypes. Transthyretin amyloidosis is a slowly progressive condition characterized by the build-up of abnormal deposits of a protein called amyloid (amyloidosis) in the body's organs and tissues. These protein deposits most frequently occur in the peripheral nervous system, which is made up of nerves connecting the brain and spinal cord to muscles and sensory cells that detect sensations such as touch, pain, heat, and sound. Protein deposits in these nerves result in a loss of sensation in the extremities (peripheral neuropathy). The autonomic nervous system, which controls involuntary body functions such as blood pressure, heart rate, and digestion, may also be affected by amyloidosis. In some cases, the brain and spinal cord (central nervous system) are affected. Other areas of amyloidosis include the heart, kidneys, eyes, and gastrointestinal tract, This analysis has been completed for multiple subtypes of Amyloidosis:

1) Cardiac transthyretin amyloidosis (ATTR-CM)

2) Neuropathic transthyretin amyloidosis (ATTR-PN)

3) Wild type transthyretin amyloidosis (wtATTR)

4) Systemic light chain amyloidosis (AL)

Findings in the ATTR-CM sub-type have been written up in a manuscript entitled Natural history, quality of life and

outcomes in cardiac ATTR amyloidosis, that has been published to the Circulation in July 2 2019

(https://www.ahajournals.org/doi/full/10.1161/CIRCULATIONAHA.118.038169 ). Findings from the analysis in the remaining two subtypes, ATTR-PN and AL, will be developed into publications to submit to peer-reviewed journals, as well as further findings from the ATTR-CM subtype.

Outputs will contain only aggregate-level data with small numbers suppressed in line with the HES analysis guide. Outputs are intended to improve the currently limited understanding of the patient journey and diagnostic pathway in amyloidosis, which could realise benefits such as a more efficient care pathway, earlier diagnosis and treatment (this is described in more detail in the expected measurable benefits section). Outputs are for peer review publications and will not be used for commercial / sales and marketing purposes.

The secondary objective of the research was to describe predictive patient characteristics to support:

1. The diagnosis of amyloidosis patients earlier in the patient pathway than they would otherwise be diagnosed

2. The flagging of high-risk patients for diagnostic tests, who otherwise may go un-diagnosed

This objective has yet to be addressed due to limited capacity from academic collaborators and discontinuation of sponsorship from industry collaborators, GlaxoSmithKline. IQVIA notes in this extension agreement that GlaxoSmithKline is no longer funding this research , and that further research will be jointly funded by IQVIA and The Royal Free Hospital.

The purpose of this extension is to secure continued access to the research dataset to:

1. Respond to any further comments from reviewers on the manuscript submitted to Circulation

2. Publish further findings from the ATTR-CM subtype beyond what is included in the first Circulation manuscript

3. Write up findings from the analysis of treatment patterns, treatment toxicity and outcomes in AL patients who received chemotherapy (due for submission September 2019)

4. Write up the findings from the analysis of diagnostic and treatment pathways in other subtypes of the disease into a manuscript for submission to a peer-reviewed journal

5. Address the secondary objective of the research: to describe predictive patient characteristics to support the flagging of patients earlier in their diagnostic pathway, or the flagging of patients who have not yet been diagnosed via the development of a predictive algorithm IQVIA expects to produce the following analyses:

1) Investigate the predictive patient characteristics within the data environment to understand if IQVIA can support the flagging of patients earlier in their diagnostic pathway or flag patients who have not yet been diagnosed via the

development of a predictive algorithm expected to be completed 12-18 months after HES data has been provided .

The target dissemination plan is as follows:

2) The applicant will submit the findings of the research to a peer review journal e.g. Rheumatology

3) The applicant will submit and present on findings at a relevant amyloidosis conference e.g. 2020 International

Symposium on Amyloidosis, in order to further the knowledge of other specialist physicians

4) Published results will be shared with the UKAAG patient advocacy group

5) Furthermore the abstracts and links to publications will be hosted on IQVIAs online bibliography which is publicly available

6) Results will also be shared with other parties where appropriate e.g. Sharing results with other NHS trusts who also manage amyloidosis patients or sharing with international centres which also diagnose and manage amyloidosis patients

The output of the algorithm generation is currently uncertain. However, any implementation would need to be conducted by or with NHS bodies, because IQVIA is working with pseudonymised data and will not seek to re-identify patients at any stage. The nature of any implementation would need to be driven by the predictive sensitivity and specificity of the algorithm. In other words, the false positive and false negative detection rate. Implementing an algorithm with a high false positive rate would lead to many people tested with very few identified, conversely if the algorithm has a high false negative rate, it will likely miss many patients who should be tested for the disease. The health economics of the algorithm and any associated intervention would need to be carefully assessed prior to any implementation. Prior to any algorithm playing a role in supporting clinical practice / being implemented it will require peer review publication and broad acceptance before any uptake could be successful.

For an algorithm with weaker predictive potential, IQVIA envisages the generation of publications in peer reviewed journals and generation of medical educational materials to use with clinical specialities who are potentially exposed to amyloidosis patients. The literature will document the methodology used and the risk factors which would help to identify amyloidosis patients earlier. These will potentially be presented at symposiums or other forums, depending on the findings.

If an algorithm with high predictive potential is generated, it could be used to create a clinical support tool for physicians to help diagnose patients, allowing the summarisation of large quantities of data in a more manageable format. This tool could support physicians by providing a risk score which they can interpret themselves to support clinician decisions.

The algorithm would be free of charge and openly available. Access methods would be dependent on the strength of the algorithm but may include presentation at seminars, publications on risk factors or a clinical support tool provided directly to physicians (subject to any relevant approvals) If no information of merit is found, the methodology utilised in the research would be documented and submitted to a peer reviewed journal. This would allow other researchers to benefit from the research efforts. In addition, the methodology would be shared via IQVIA's online bibliography and which is publicly available. In all summaries, any data used would be aggregated with small numbers suppressed in line with the HES Analysis Guide.

No organisation on the clinical interpretation group will have the ability to suppress the dissemination of findings or outputs from this work. None of these outputs are linked to PhD studies.

Based on the data provided for the previous agreement, IQVIA has developed a full set of analysis regarding the NAC patient pathway for Amyloidosis which have been discussed with both the NAC (the academic collaborators) and GSK (an industry partner who previously, but no longer, sponsored this research). As a result of these

discussions, there is a considerably better understanding of the Amyloidosis patient pathway, including per subtype.

For example, for ATTR-CM patients treated at the NAC:

- There is a substantial delay in diagnosis following onset of symptoms, with patients using hospital services (either as an inpatient, outpatient or Accident & Emergency visit) a mean of 19.9 times during the 3 years before diagnosis; diagnosis of the wild-type form of ATTR-CM was delayed more than 4 years after onset of cardiac

symptoms in 42% of cases

- Hereditary cardiac amyloidosis patients with a certain genetic mutation (V122I) were more impaired functionally and had worse measures of cardiac disease at the time of diagnosis, and poorer survival compared to the other sub-groups (such as patients carrying a different genetic mutation, and patients with wild-type form of the disease)

- Analysis of the diagnostic and treatment pathways for different amyloidosis subtypes has also been completed after HES data was provided, leading to improved knowledge and patient benefits.

All parties now have a better understanding of treatments that occur outside of the NAC and patient outcomes. A set of papers from the outputs of the analysis have been already published as mentioned above or will be developed into publication and published in research journals, with one manuscript, entitled Natural history, quality of life and outcomes in cardiac ATTR amyloidosis, already published in the journal Circulation. Through publication, the results have been shared across the scientific community, increasing the body of knowledge on Amyloidosis disease and benefiting Amyloidosis patients with potential future treatment innovations.