Alternative to outerjoin for large table concatenation?

Hi MATLAB'ers,

I am looking for tips on how to speed up vertical concatenation of tables. Specifically, how to identify and append missing columns to each table before concatenation.

In my case, I have 2000 tables stored in a cell array and ~15000 unique columns (VariableNames). Each table has a randomly-ordered subset of between 10-3000 of these columns and, importantly, a column containing unique sample identifiers.

One way of tackling this is to simply call outerjoin in a loop. Outerjoin normalizes the column headers for you, but spends an inordinate amount of time in the joinInnerOuter.m and defaultarrayLike.m subfunctions.

    MergeTable = AllTablesCellArray{1};

    for nt = 2:length(MutTables)
        MergeTable = outerjoin(MergeTable,AllTablesCellArray{nt},'MergeKeys',true);
    end

Another strategy is to pad the individual tables with the extra (missing) columns then vertcat. This works, but is slower. On the other hand, this code can use a parfor loop.

    AllVars   = cellfun(@(x) x.Properties.VariableNames,AllTablesCellArray,'UniformOutput',false);
    UniqueVars= unique([AllVars{:}],'stable');

        for nt = 1:length(AllTablesCellArray)
            MissingVars = UniqueVars(~ismember(UniqueVars,AllVars{nt}))'; 

            if ~isempty(MissingVars)
                AllTablesCellArray{nt}{:,MissingVars}  = repmat({''},height(AllTablesCellArray{nt}),length(MissingVars));
            end
        end

    MergeTable = vertcat(AllTablesCellArray{:});

I am hoping that my brain is just fried and that I am missing something obvious. My hope is to avoid converting each table to a cell array, but this could be a good way to go.

Thanks!

-Jeff