Abbreviated Smell Test to Identify Patients with PD/DLB

Code for analysis in the manuscript entitled "Development of a Simplified Smell Test to Identify Patients with Typical Parkinson’s as Informed by Multiple Cohorts, Machine Learning and External Validation".

medRxiv preprint doi: https://doi.org/10.1101/2024.08.09.24311696

Data files

Datasets of the 3 cohorts used in this study can be accessed via zenodo (link to be updated) upon request.

Note: please place the data files within the same working directory after downloading them from zenodo.

1. Data dictionary.xlsx
2. Ottawa_cut.csv: Ottawa (PREDIGT) Trial (baseline)
3. PROBE_cut.csv: Prognostic Biomarkers in Parkinson’s Disease Study (PROBE; baseline)
4. DeNoPa_cut.csv: “De Novo Parkinson disease study” (DeNoPa; baseline, 48-month, and 72-month follow-up visits)

Note: the public accessible data files have gone through further de-identification steps to protect the identity of all study participants. If you wish to reproduce all the results in the manuscript, please contact the corresponding authors or cohort PIs to access the original data sets. The de-identification steps include:

1. Transform continuous age to categorical age_cat: This will affect the results of Table 1, Supplemental Table 3, and Figure 7(b) in the medRxiv version.
2. Remove disease.duration: This will affect the results of Table 1.
3. Remove race, ethnic: These variables were not directly used to generate any results in the manuscript, but were used to describe the race/ethnicity characteristics of the study cohorts.

Supportive data

Included in the Supportive data folder in this repo:

1. diagnosis colors.csv: colors representing each diagnostic groups for plotting.
2. Published rankings.csv: 8 previously published scent rankings, 4 for SST-ID, 4 for UPSIT.
3. scent_shared.csv: 11 scents that are shared by SST-ID and UPSIT. (Note: this file is generated by 4. Combine UPSIT and SST-ID.qmd, but is also required by 1. Tables, distributions, ROCs.qmd for plotting. So I included it here to avoid running error.)
4. SST_ID_option.csv: Options provided for each SST-ID scent, used for generating ICCs.
5. UPSIT_key.csv: answer keys of UPSIT.
6. UPSIT_option.csv: Options provided for each UPSIT scent, used for generating ICCs.

Code files

Important: This project used TestGardener version 3.2.6. TestGardener has had a major update recently (version 3.3.3) which includes various changes that are incompatible with code here. We will update the repo to accommodate the changes, but until that completes, please use TestGardener version 3.2.6 to run this code. You can find it here.

1. 1. Tables, distributions, ROCs.qmd
   1. Code for
      • Table 1: demographic and diagnostic characteristics of the 3 cohorts
      • Table 2: AUC, sensitivity and specificity of SST-ID and UPSIT
      • Supplemental Table 2: relationship between SST-ID/UPSIT scores with age, sex, and diagnostic groups
      • Figure 2: distribution and AUC values of SST-ID and UPSIT
      • Supplemental Figure 1: distribution and AUC values of SST-TH and SST-DS
   2. Need to read
      • Ottawa_cut.csv
      • PROBE_cut.csv
      • DeNoPa_cut.csv
      • Supportive data/diagnosis colors.csv
2. 2. Abbreviated Smell Test SST-ID.qmd
   1. Code for
      • Figure 3 (a,b): rankings of SST-ID scents, percentages of correct scent identification in each group
      • Supplemental Figure 2 (a): Percentage differences of correct scent identification between HC and PD/DLB groups (% HC - % PD/DLB) in DeNoPa (SST-ID)
      • Figure 4 (a): comparing SST-ID rankings of this study with 4 previously published ones
      • Figure 6 (a)-(c): validation of the SST-ID subsets
      • Figure 5, Supplemental Figure 3, Supplemental Figure 4: ICCs of SST-ID scents
   2. Need to read
      • DeNoPa_cut.csv
      • Supportive data/Published rankings.csv
      • Supportive data/scent_shared.csv
      • Supportive data/SST_ID_option.csv
   3. Generate files: the intermediate results will be saved to a folder called Generated files. Note: you need to create such folder within your working directory.
      • SST_ID_rankings.csv: rankings of SST-ID scents
      • df_sex_D_M.rds, df_sex_D_F.rds, df_sex_D.rds: relationship between scent identification and sex of SST-ID in DeNoPa
      • df_fine_D.rds, df_age_D.rds: relationship between scent identification and age of SST-ID in DeNoPa Note: These cannot be reproduced if you are using data from zenodo
3. 3. Abbreviated Smell Test UPSIT.qmd
   1. Code for
      • Figure 3 (c,d): rankings of UPSIT scents, percentages of correct scent identification in each group
      • Supplemental Figure 2 (b): Percentage differences of correct scent identification between HC and PD/DLB groups (% HC - % PD/DLB) in Ottawa Trial (UPSIT)
      • Figure 4 (b): comparing UPSIT rankings of this study with 4 previously published ones
      • Figure 6 (d,e): validation of the UPSIT subsets
      • Figure 5, Supplemental Figure 3, Supplemental Figure 5: ICCs of UPSIT scents
   2. Need to read
      • Ottawa_cut.csv
      • PROBE_cut.csv
      • Supportive data/diagnosis colors.csv
      • Supportive data/Published rankings.csv
      • Supportive data/scent_shared.csv
      • Supportive data/UPSIT_key.csv
      • Supportive data/UPSIT_option.csv
   3. Generate files: the intermediate results will be saved to a folder called Generated files. Note: you need to create such folder within your working directory.
      • UPSIT_rankings.csv: rankings of UPSIT scents
      • df_sex_O_M.rds, df_sex_O_F.rds, df_sex_O.rds: relationship between scent identification and sex of UPSIT in Ottawa trial
      • df_fine_O.rds, df_age_O.rds: relationship between scent identification and age of UPSIT in Ottawa trial Note: These cannot be reproduced if you are using data from zenodo
      • df_sex_P_M.rds, df_sex_P_F.rds, df_sex_P.rds: relationship between scent identification and sex of UPSIT in PROBE
      • df_fine_P.rds, df_age_P.rds: relationship between scent identification and age of UPSIT in PROBE Note: These cannot be reproduced if you are using data from zenodo
4. 4. Combine UPSIT and SST-ID.qmd
   1. Code for
      • Table 3: rankings of the 11 scents shared by SST-ID and UPSIT
      • Table 4: performance of the 7-scent abbreviated smell test
      • Figure 7: relationship of scent identification with sex and age
   2. Need to read
      • Ottawa_cut.csv
      • PROBE_cut.csv
      • DeNoPa_cut.csv
      • Supportive data/diagnosis colors.csv
      • Intermediate results generated by 2. Abbreviated Smell Test SST-ID.qmd and 3. Abbreviated Smell Test UPSIT.qmd
   3. Generate files: the intermediate results will be saved to a folder called Generated files. Note: you need to create such folder within your working directory.
      • scent_shared.csv: 11 scents that are shared by SST-ID and UPSIT.

Supportive function

Supportive functions are placed in the R folder. These functions need to be sourced in 2. Abbreviated Smell Test SST-ID.qmd and 4. Combine UPSIT and SST-ID.qmd.

1. itemAUC.R: calculate each scent's AUC values and rank them, using cross-validation
2. subsetAUC.R: calculate each subset's AUC values.
3. Modified TestGardener functions to generate ICCs: make.dataList.R, Wbinsmth.R, eval_surp.R, ICC.plot.R