semantically, it is more recommended to use your way of writing, which is simple and intuitive. In fact, your writing method can be simplified to

 let str = arr.filter(item => item.status).map(item => item.id).join(',');

however, this method of writing has been cycled twice. In pursuit of efficiency, you can try to change the following method of writing only once in a loop, but the code is difficult to understand

let str = arr.reduce((_arr, item) => item.status && _arr.push(item.id) ? _arr : _arr, []).join(',');

if you optimize manually, you can merge the loop.
you can also rely on the lazy evaluation feature of lodash to let lodash optimize it automatically

let str=_.chain(arr).filter('status').map('id').join(',').value()

let arr = [];
for (let i = 1; i < 10000000; iPP) {
    arr.push({id: i, status: Math.random() > 0.5});
}

console.time('time1');
let arr2 = arr.filter(item => item.status);
let arr3 = arr2.map(item => item.id);
var str = arr3.join(',');
console.timeEnd('time1');

const _ = require('lodash');
const calculateChain = _.chain(arr).filter('status').map('id').join(',');
console.time('time2');
var str = calculateChain.value();
console.timeEnd('time2');

console.time('time3');
var str = arr.reduce((_arr, item) => item.status && _arr.push(item.id) ? _arr : _arr, []).join(',');
console.timeEnd('time3');

it is the fastest to use lodash in actual measurement. It should be that join is also an o (n) operation, and there is room for optimization